JP5036626B2 - Image encoding apparatus and image decoding apparatus - Google Patents

Description

  The present invention relates to an image encoding device that generates encoded data by quantizing the gradation level of each pixel indicated by image data output from an image sensor, and to decode encoded data output from the image encoding device. The present invention relates to an image decoding apparatus that generates image data.

In the image encoding apparatus of the FBTC encoding method, an image is divided into M × N blocks (M and N are natural numbers), and for each block, a pixel reference level LA in which pixel values in the block are averaged, and pixels A pixel value fluctuation range LD indicating a difference between the maximum value and the minimum value is obtained.
Then, a quantization threshold for quantizing the pixel value of each pixel is calculated from the pixel reference level LA and the pixel value fluctuation width LD, and the pixel value of each pixel is compared with the quantization threshold. Is converted into a quantized value to compress the image.

For example, Patent Document 1 below discloses a configuration in which an internal division ratio for obtaining a quantization threshold value for each block is made variable so that the internal division ratio can be designated from the outside.
In addition, a configuration in which an internal ratio is selected for each block and information indicating the internal ratio is encoded as additional information is also disclosed.
Furthermore, a configuration is also disclosed in which an internal ratio is selected for each block according to the distribution of image data.

However, in the image encoding device disclosed in Patent Document 1 below, the encoding size of each block is fixed.
This is one of the characteristics of the FBTC coding system. However, in general image characteristics, a region with a small amount of information with a small variation in pixel value and a region with a large amount of information with a large variation in pixel value are mixed. is doing.
For this reason, if the entire image is fixed at the same encoding size, the difference in image quality may be large between a region with a large amount of information and a region with a small amount of information.

In addition, even when a natural image region such as a camera image and a region such as a map or a drawing are mixed in the expanded image, the natural image region contains a large amount of information, so the image quality is greatly deteriorated. On the other hand, the image quality in the drawing area is reduced.
For this reason, the consistency of image quality may not be maintained for the entire display image.

Japanese Patent Laid-Open No. 6-292027 (paragraph numbers [0015] to [0016], FIG. 1)

  Since the conventional image coding apparatus is configured as described above, the entire image is fixed at the same coding size, the pixel value variation is small, the information amount is small, and the pixel value variation is large. In the case of an image in which a region with a large amount of information is mixed, there is a problem that a difference in image quality may increase between a region with a large amount of information and a region with a small amount of information.

The present invention has been made to solve the above-described problems, and an object of the present invention is to obtain an image encoding apparatus capable of improving the encoding efficiency and the image quality of the entire image.
It is another object of the present invention to provide an image decoding apparatus that can generate a homogeneous image with little difference in image quality over the entire image.

The image encoding apparatus according to the present invention is based on a reference of the gradation level in each block from the maximum value and the minimum value detected by the signal level range detection means, corresponding to a plurality of data quantization methods prepared in advance. A plurality of encoding preparation means for calculating a level and a fluctuation range, and calculating a quantization threshold of a gradation level in each block from the reference level and fluctuation width calculated by the encoding preparation means, and the reference level and fluctuation width A plurality of quantization threshold calculation means for calculating the gradation level of each pixel when a unique code assigned as quantized data is decoded, and the gradation level and quantization threshold calculation of each pixel in each block The quantization threshold calculated by the means is compared to identify a unique code corresponding to the gradation level of each pixel, and the unique code is converted into the quantization data. A plurality of quantization means to be assigned to each pixel, and the gradation level of each pixel when the unique code calculated by the quantization threshold value calculation means is decoded and before the quantization data is assigned by the quantization means A plurality of quantization error integration means for integrating the absolute value of the error with the gradation level of each pixel for each block, and among the quantized data assigned by the plurality of quantization means for each block , Selecting the quantized data with the smallest data amount under the condition that the absolute value of the error accumulated by the quantization error accumulating means does not exceed the allowable threshold, and quantizing the quantized data and the quantized data It includes a method identification information, and encoded data generating means for generating a coded data consisting of the calculated reference level and fluctuation width by the encoding preparation means indicating how the encoded data The generating means calculates the target code amount per block of the unprocessed block from the target code amount per image frame, the accumulated code amount of the processed blocks, and the number of unprocessed blocks. The allowable threshold value is determined from the target code amount per block of the processing block .

According to the present invention, as corresponding to a plurality of data quantization methods prepared in advance, the reference level and fluctuation range of the gradation level in each block from the maximum value and the minimum value detected by the signal level range detection means A plurality of encoding preparation means for calculating the threshold value, and the quantization threshold of the gradation level in each block from the reference level and the fluctuation range calculated by the encoding preparation means, and the quantization level from the reference level and fluctuation width Calculated by a plurality of quantization threshold calculation means for calculating the gradation level of each pixel when a unique code assigned as data is decoded, and the gradation level of each pixel in each block and the quantization threshold calculation means The unique quantization code corresponding to the gradation level of each pixel is identified, and the unique code is used as quantized data for each pixel. A plurality of quantization means to be assigned, a gradation level of each pixel when the unique code calculated by the quantization threshold calculation means is decoded, and a level of each pixel before the quantization data is assigned by the quantization means A plurality of quantization error integration means for integrating the absolute value of the error with the key level for each block, and among the quantized data allocated by the plurality of quantization means for each block, the quantization error Method identification information indicating the quantized data and the quantizing method related to the quantized data by selecting the quantized data with the smallest data amount under the condition that the absolute value of the error accumulated by the accumulating means does not exceed the allowable threshold , and includes a coded data generating means for generating a coded data consisting of the calculated reference level and fluctuation width by the encoding preparing means, the encoded data generating means, image 1 frame The target code amount per unprocessed block is calculated from the target code amount per block, the accumulated code amount of processed blocks, and the number of unprocessed blocks, and one block of unprocessed blocks Since the permissible threshold is determined from the target code amount per hit, there is an effect that it is possible to improve the encoding efficiency and the image quality of the entire image.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing an image encoding apparatus according to Embodiment 1 of the present invention. In the figure, an image data dividing unit 1 outputs image data output from an image sensor M × N (M and N are natural numbers). The process of dividing into blocks is performed. The image data dividing unit 1 constitutes image data dividing means.
The signal level range detection unit 2 refers to the gradation level Xij (for example, 1 ≦ i ≦ 4, 1 ≦ j ≦ 4) of each pixel indicated by the image data divided by the image data division unit 1, in each block. A process of detecting the maximum value Lmax and the minimum value Lmin of the gradation level is performed. The signal level range detector 2 constitutes a signal level range detection means.

The encoding unit 3 quantizes the gradation level of each pixel in each block using the maximum value Lmax and the minimum value Lmin of the gradation level in each block detected by the signal level range detection unit 2, and encodes the encoded data. Implement the process to create. Note that the encoding unit 3 constitutes an encoding unit.
The quantization unit 4a performs a process of quantizing the gradation level of each pixel in each block by a data quantization method a prepared in advance.
The quantization unit 4b performs a process of quantizing the gradation level of each pixel in each block by a data quantization method b prepared in advance.
The quantization unit 4c performs a process of quantizing the gradation level of each pixel in each block by a data quantization method c prepared in advance.
In the first embodiment, three quantizers 4a to 4c are mounted, but this is only an example, and four or more quantizers may be mounted.

The encoding preparation unit 5a of the quantization unit 4a calculates the pixel reference level LAa and the pixel value variation width LDa of the gradation level in each block from the maximum value Lmax and the minimum value Lmin detected by the signal level range detection unit 2. To implement.
The quantization threshold value calculation unit 6a calculates a gradation level quantization threshold value La1 in each block from the pixel reference level LAa and the pixel value variation width LDa calculated by the encoding preparation unit 5a, and also calculates the pixel reference level LAa and the pixel value. A process of calculating the gradation levels Qa1 to Qa2 of each pixel when a unique code assigned as quantized data by the quantization processor 7a is decoded from the variation width LDa is performed.
The quantization processing unit 7a compares the gradation level Xij of each pixel in each block with the quantization threshold value La1 calculated by the quantization threshold value calculation unit 6a, and assigns a unique code corresponding to the gradation level Xij of each pixel. A process of identifying and assigning the unique code to each pixel as quantized data φaij (for example, 1 ≦ i ≦ 4, 1 ≦ j ≦ 4) is performed.
The quantization error accumulating unit 8a has the gradation levels Qa1 to Qa2 of each pixel when the unique code calculated by the quantization threshold calculating unit 6a is decoded and before the quantized data is allocated by the quantization processing unit 7a. The process of integrating the absolute value of the error δaij (for example, 1 ≦ i ≦ 4, 1 ≦ j ≦ 4) with the gradation level Xij of each pixel for each block is performed.

The encoding preparation unit 5b of the quantization unit 4b calculates the pixel reference level LAb and the pixel value variation width LDb of the gradation level in each block from the maximum value Lmax and the minimum value Lmin detected by the signal level range detection unit 2. To implement.
The quantization threshold calculation unit 6b calculates the gradation threshold quantization thresholds Lb1 to Lb3 in each block from the pixel reference level LAb and the pixel value variation width LDb calculated by the encoding preparation unit 5b, and the pixel reference level LAb and A process of calculating gradation levels Qb1 to Qb4 of each pixel when a unique code assigned as quantized data by the quantization processing unit 7b is decoded from the pixel value variation width LDb is performed.
The quantization processing unit 7b compares the gradation level Xij of each pixel in each block with the quantization thresholds Lb1 to Lb3 calculated by the quantization threshold calculation unit 6b, and determines the uniqueness corresponding to the gradation level Xij of each pixel. A process is performed in which a code is specified and the unique code is assigned to each pixel as quantized data φbij (for example, 1 ≦ i ≦ 4, 1 ≦ j ≦ 4).
The quantization error accumulating unit 8b has the gradation levels Qb1 to Qb4 of each pixel when the unique code calculated by the quantization threshold calculating unit 6b is decoded and before the quantized data is allocated by the quantization processing unit 7b. A process of integrating the absolute value of an error δbij (for example, 1 ≦ i ≦ 4, 1 ≦ j ≦ 4) from the gradation level Xij of each pixel for each block is performed.

The encoding preparation unit 5c of the quantization unit 4c calculates the pixel reference level LAc and the pixel value variation width LDc of the gradation level in each block from the maximum value Lmax and the minimum value Lmin detected by the signal level range detection unit 2. To implement.
The quantization threshold calculation unit 6c calculates the gradation threshold quantization thresholds Lc1 to Lc7 in each block from the pixel reference level LAc and the pixel value variation width LDc calculated by the encoding preparation unit 5c, and the pixel reference level LAc and A process of calculating gradation levels Qc1 to Qc8 of each pixel when a unique code assigned as quantized data is decoded by the quantization processing unit 7c from the pixel value variation width LDc is performed.
The quantization processing unit 7c compares the gradation level Xij of each pixel in each block with the quantization threshold values LC1 to Lc7 calculated by the quantization threshold value calculation unit 6c, and determines the uniqueness corresponding to the gradation level Xij of each pixel. A process is performed in which a code is specified and the unique code is assigned to each pixel as quantized data φcij (for example, 1 ≦ i ≦ 4, 1 ≦ j ≦ 4).
The quantization error accumulating unit 8c has gradation levels Qc1 to Qc8 of each pixel when the unique code calculated by the quantization threshold calculating unit 6c is decoded and before the quantized data is allocated by the quantization processing unit 7c. The process of integrating the absolute value of the error δcij (for example, 1 ≦ i ≦ 4, 1 ≦ j ≦ 4) from the gradation level Xij of each pixel for each block is performed.

The encoding preparation units 5a, 5b, and 5c constitute an encoding preparation unit, and the quantization threshold calculation units 6a, 6b, and 6c constitute a quantization threshold calculation unit.
The quantization processing units 7a, 7b, and 7c constitute quantization means, and the quantization error integration units 8a, 8b, and 8c constitute quantization error integration means.

The encoded data creating unit 9 is an error accumulated by the quantization error accumulating units 8a, 8b, and 8c among the quantized data φaij, φbij, and φcij assigned by the quantization processing units 7a, 7b, and 7c for each block. Under the condition that the absolute values of δaij, δbij, and δcij do not exceed the allowable threshold value ΔT, the quantized data φnij (n = “a” or “b” or “c”) having the smallest data amount is selected and the quantum Encoded data consisting of the encoded data φnij, a flag (method identification information) indicating the data quantization method n related to the quantized data φnij, the pixel reference level LAn and the pixel value variation width LDn calculated by the encoding preparation unit 5n Implement the process to create. The encoded data creation unit 9 constitutes encoded data creation means.
FIG. 2 is a flowchart showing the processing contents of the image coding apparatus according to Embodiment 1 of the present invention.

Next, the operation will be described.
When receiving image data from the image sensor, the image data dividing unit 1 divides the image data into M × N blocks (M and N are natural numbers) (step ST1).
In the first embodiment, for convenience of explanation, each block is assumed to be composed of 16 pixels of 4 horizontal pixels and 4 vertical pixels.

When the image data dividing unit 1 divides the image data into M × N blocks, the signal level range detecting unit 2 selects one unprocessed block that has not yet been encoded from the M × N blocks. Are selected (step ST2).
When the unprocessed block is selected, the signal level range detection unit 2 refers to the gradation level of each pixel in the block and detects the maximum value Lmax and the minimum value Lmin of the gradation level in the block (step ST3). ).

When the signal level range detection unit 2 detects the maximum value Lmax and the minimum value Lmin of the gradation level, the encoding preparation unit 5a of the quantization unit 4a uses the calculation method related to the data quantization method a to calculate the gradation level. From the maximum value Lmax and the minimum value Lmin, the pixel reference level LAa and the pixel value variation width LDa of the gradation level in the block are calculated (step ST4a).
Also, the encoding preparation unit 5b of the quantization unit 4b uses the calculation method related to the data quantization method b when the signal level range detection unit 2 detects the maximum value Lmax and the minimum value Lmin of the gradation level. From the maximum value Lmax and the minimum value Lmin of the level, the pixel reference level LAb and the pixel value variation width LDb of the gradation level in the block are calculated (step ST4b).
Also, the encoding preparation unit 5c of the quantization unit 4c uses the calculation method related to the data quantization method c when the signal level range detection unit 2 detects the maximum value Lmax and the minimum value Lmin of the gradation level. From the maximum value Lmax and the minimum value Lmin of the level, the pixel reference level LAc and the pixel value variation width LDc of the gradation level in the block are calculated (step ST4c).
Hereinafter, a calculation method of the pixel reference level LAn and the pixel value variation width LDn in the encoding preparation unit 5n (n = “a” or “b” or “c”) will be specifically described.

First, as shown in the following formulas (1) and (2), the encoding preparation unit 5n calculates thresholds P1n and P2n for determining a reference pixel when calculating the pixel reference level LAn and the pixel value variation width LDn. To do.
P1n = (K1n × Lmax + K2n × Lmin) / (K1n + K2n) (1)
P2n = (K2n × Lmax + K1n × Lmin) / (K1n + K2n) (2)
Here, the threshold values P1n and P2n are calculated as values obtained by weighted averaging of the maximum value Lmax and the minimum value Lmin of the gradation level. At this time, the coefficients K1n and K2n applied to the weighted average are the gradation values of the image. A value that minimizes the quantization error can be selected according to the level distribution state, the number of bits assigned when converting the gradation level of each pixel, and the characteristics of the image of the system to which the present technology is applied. is there.

In the first embodiment, for example, the data quantization method a has a 1-bit 2 gradation number assigned to a code, the data quantization method b has a 2-bit 4 gradation number assigned to a code, and the data quantization method c has a code. The number of gradations to be assigned to is 3 bits and 8 gradations, and the values of the coefficients K1n and K2n are as shown in FIG.
In addition, regarding the gradation level Xij (1 ≦ i ≦ 4, 1 ≦ j ≦ 4) of each pixel of the block, the average value of gradation levels of pixels satisfying Xij ≦ P1n is defined as Qmin_n.
Further, an average gradation level value of pixels satisfying Xij> P2n is set to Qmax_n.

The encoding preparation unit 5n calculates the pixel reference level LAn and the pixel value variation width LDn by substituting the gradation level average values Qmin_n and Qmax_n into the following equations (3) and (4).
LAn = (Qmin_n + Qmax_n) / 2 (3)
LDn = (Qmax_n−Qmin_n) (4)

When the encoding preparation unit 5a calculates the pixel reference level LAa and the pixel value variation width LDa, the quantization threshold calculation unit 6a of the quantization unit 4a determines the gradation in the block from the pixel reference level LAa and the pixel value variation width LDa. The quantization threshold value La1 of the level is calculated, and from the pixel reference level LAa and the pixel value variation width LDa, each pixel level when a unique code assigned as quantized data by the quantization processing unit 7a is decoded. Key levels (decoding levels) Qa1 to Qa2 are calculated (step ST5a).
In addition, when the encoding preparation unit 5b calculates the pixel reference level LAb and the pixel value variation width LDb, the quantization threshold calculation unit 6b of the quantization unit 4b uses the pixel reference level LAb and the pixel value variation width LDb in the block. When the quantization thresholds Lb1 to Lb3 of the gradation level are calculated, and a unique code assigned as quantized data by the quantization processing unit 7b is decoded from the pixel reference level LAb and the pixel value fluctuation width LDb. The gradation levels (decoding levels) Qb1 to Qb4 of each pixel are calculated (step ST5b).
Further, the quantization threshold calculation unit 6c of the quantization unit 4c calculates the pixel reference level LAc and the pixel value variation width LDc from the pixel reference level LAc and the pixel value variation width LDc when the encoding preparation unit 5c calculates the pixel reference level LAc and the pixel value variation width LDc. While calculating the quantization thresholds Lc1 to Lc7 of the gradation level, each of the unique codes assigned as quantized data by the quantization processing unit 7c from the pixel reference level LAc and the pixel value fluctuation width LDc is decoded. Pixel gradation levels (decoding levels) Qc1 to Qc8 are calculated (step ST5c).

The number of quantization thresholds calculated by the quantization threshold calculation units 6a, 6b, and 6c needs to correspond to the number of bits assigned to the code for converting the gradation level.
The data quantization method a is 1 bit 2 gradations for the number of gradations assigned to the code, the data quantization method b is 2 bits 4 gradations for the number of gradations assigned to the code, and the data quantization method c is the 3 bits 8 gradations assigned to the code. Therefore, the quantization threshold value calculation unit 6a corresponding to the data quantization method a has one quantization threshold value, the quantization threshold value calculation unit 6b corresponding to the data quantization method b has three quantization threshold values, The quantization threshold calculation unit 6c corresponding to the quantization method c calculates seven quantization thresholds.

The calculation formulas of the quantization threshold values by the quantization threshold value calculation units 6a, 6b, and 6c are as the following formulas (5) to (15).
However, the calculation method of the quantization threshold value by the formulas (5) to (15) is an example, and the quantization error is as small as possible according to the gradation level distribution state of the image and the image characteristics of the system to which the present technology is applied. And the calculation method is not necessarily limited to the equations (5) to (15) listed here.

[In the case of data quantization method a]
La1 = LAa (5)
[In the case of data quantization method b]
Lb1 = LAb−LDb / 4 (6)
Lb2 = LAb (7)
Lb3 = LAb + LDb / 4 (8)
[In the case of data quantization method c]
Lc1 = LAc−LDc × 3/8 (9)
Lc2 = LAc−LDc × 1/4 (10)
Lc3 = LAc−LDc × 1/8 (11)
Lc4 = LAc (12)
Lc5 = LAc + LDc × 1/8 (13)
Lc6 = LAc + LDc × 1/4 (14)
Lc7 = LAc + LDc × 3/8 (15)

Further, as the gradation level of each pixel when a unique code is decoded, it is necessary to obtain 2 for the data quantization method a, 4 for the data quantization method b, and 8 for the data quantization method c. .
Expressions for calculating the gradation levels of the respective pixels by the quantization threshold calculation units 6a, 6b, and 6c are as shown in the following expressions (16) to (29).
However, the formulas for calculating the gradation levels according to the equations (16) to (29) are only examples. When the formulas (5) to (15), which are quantization threshold calculation formulas, are changed, the formulas (16) to (29) are also changed to correspond to the change.

[In the case of data quantization method a]
Qa1 = LAa−LDa / 2 (16)
Qa2 = LAa + LDa / 2 (17)
[In the case of data quantization method b]
Qb1 = LAb−LDb / 2 (18)
Qb2 = LAb−LDb / 8 (19)
Qb3 = LAb + LDb / 8 (20)
Qb4 = LAb + LDb / 2 (21)
[In the case of data quantization method c]
Qc1 = LAc−LDc / 2 (22)
Qc2 = LAc−LDc × 5/16 (23)
Qc3 = LAc−LDc × 3/16 (24)
Qc4 = LAc-LDc / 16 (25)
Qc5 = LAc + LDc / 16 (26)
Qc6 = LAc + LDc × 3/16 (27)
Qc7 = LAc + LDc × 5/16 (28)
Qc8 = LAc + LDc / 2 (29)

The quantization processing units 7a, 7b, and 7c of the quantization units 4a, 4b, and 4c are the quantization threshold value calculation units 6a, 6b, and 6c, and the quantization level quantization threshold values La1, Lb1 to Lb3, and Lc1 to Lc7 in the block. Is calculated, an unprocessed pixel in the block is selected (step ST6).
When the quantization processing unit 7a of the quantization unit 4a selects an unprocessed pixel, it compares the gradation level Xij of that pixel with the quantization threshold value La1 calculated by the quantization threshold value calculation unit 6a, A unique code corresponding to the gradation level Xij is specified, and the unique code is assigned to the pixel as quantized data φaij (step ST7a).
That is, the quantization processing unit 7a determines which region the gradation level Xij of the pixel belongs to among the two regions divided by the quantization threshold value La1, and quantizes the unique code corresponding to the region to which the quantization processing unit 7a belongs. A process of assigning the pixel as the converted data φaij is performed.

When the quantization processing unit 7b of the quantization unit 4b selects an unprocessed pixel, the gradation level Xij of the pixel is compared with the quantization threshold values Lb1 to Lb3 calculated by the quantization threshold value calculation unit 6b. A unique code corresponding to the gradation level Xij of the pixel is specified, and the unique code is assigned to the pixel as the quantized data φbij (step ST7b).
That is, the quantization processing unit 7b determines which region the gradation level Xij of the pixel belongs to among the four regions divided by the quantization thresholds Lb1 to Lb3, and assigns a unique code corresponding to the region to which it belongs. A process of assigning to the pixel as quantized data φbij is performed.

When selecting an unprocessed pixel, the quantization processing unit 7c of the quantization unit 4c compares the gradation level Xij of the pixel with the quantization thresholds Lc1 to Lc7 calculated by the quantization threshold calculation unit 6c, A unique code corresponding to the gradation level Xij of the pixel is specified, and the unique code is assigned to the pixel as quantized data φcij (step ST7c).
That is, the quantization processing unit 7c determines which region the gradation level Xij of the pixel belongs to among the eight regions divided by the quantization thresholds Lc1 to Lc7, and assigns a unique code corresponding to the region to which it belongs. A process of assigning to the pixel as quantized data φcij is performed.

Specific code assignment methods by the quantization processing units 7a, 7b, and 7c are exemplified by the following equations (30) to (43).
[In the case of data quantization method a]
Xij ≦ La1 → φaij = 0 (binary number) (30)
La1 <Xij → φaij = 1 (binary number) (31)
[In the case of data quantization method b]
Xij ≦ Lb1 → φbij = 00 (binary number) (32)
Lb1 <Xij ≦ Lb2 → φbij = 01 (binary number) (33)
Lb2 <Xij ≦ Lb3 → φbij = 10 (binary number) (34)
Lb3 <Xij → φbij = 11 (binary number) (35)
[In the case of data quantization method c]
Xij ≦ Lc1 → φcij = 000 (binary number) (36)
Lc1 <Xij ≦ Lc2 → φcij = 001 (binary number) (37)
Lc2 <Xij ≦ Lc3 → φcij = 010 (binary number) (38)
Lc3 <Xij ≦ Lc4 → φcij = 011 (binary number) (39)
Lc4 <Xij ≦ Lc5 → φcij = 100 (binary number) (40)
Lc5 <Xij ≦ Lc6 → φcij = 101 (binary number) (41)
Lc6 <Xij ≦ Lc7 → φcij = 110 (binary number) (42)
Lc7 <Xij → φcij = 111 (binary number) (43)

The quantization error accumulation unit 8a of the quantization unit 4a uses the gradation levels Qa1 to Qa2 of each pixel when the unique code calculated by the quantization threshold calculation unit 6a is decoded, and the quantization processing unit 7a. An error δaij with respect to the gradation level Xij of each pixel before the quantized data φaij is assigned is calculated, and the absolute value of the error δaij is integrated in the block (step ST8a).
Further, the quantization error integrating unit 8b of the quantization unit 4b includes the gradation levels Qb1 to Qb4 of each pixel when the unique code calculated by the quantization threshold calculation unit 6b is decoded, and a quantization processing unit. The error δbij with respect to the gradation level Xij of each pixel before the quantization data φbij is assigned is calculated by 7b, and the absolute value of the error δbij is integrated in the block (step ST8b).
In addition, the quantization error accumulation unit 8c of the quantization unit 4c includes gradation levels Qc1 to Qc8 of each pixel when the unique code calculated by the quantization threshold calculation unit 6c is decoded, and a quantization processing unit. The error δcij with the gradation level Xij of each pixel before the quantization data φcij is assigned is calculated by 7c, and the absolute value of the error δcij is integrated in the block (step ST8c).

The calculation formula of the error δnij by the quantization error integration sections 8a, 8b, and 8c and the calculation formula of the integration value Δn (n = “a” or “b” or “c”) are the following formulas (44) to (60). It is as follows.
However, the value of the integrated value Δn is initialized at the start of processing of a new block, and the error δnij of each pixel is accumulated during the block processing.

[In the case of data quantization method a]
φaij = 0 (binary number) → δaij = | Xij−Qa1 | (44)
φaij = 1 (binary number) → δaij = | Xij−Qa2 | (45)
Δa = Σδaij (1 ≦ i ≦ 4, 1 ≦ j ≦ 4) (46)
[In the case of data quantization method b]
φbij = 00 (binary number) → δbij = | Xij−Qb1 | (47)
φbij = 01 (binary number) → δbij = | Xij−Qb2 | (48)
φbij = 10 (binary number) → δbij = | Xij−Qb3 | (49)
φbij = 11 (binary number) → δbij = | Xij−Qb4 | (50)
Δb = Σδbij (1 ≦ i ≦ 4, 1 ≦ j ≦ 4) (51)
[In the case of data quantization method c]
φcij = 000 (binary number) → δcij = | Xij−Qc1 | (52)
φcij = 001 (binary number) → δcij = | Xij−Qc2 | (53)
φcij = 010 (binary number) → δcij = | Xij−Qc3 | (54)
φcij = 011 (binary number) → δcij = | Xij−Qc4 | (55)
φcij = 100 (binary number) → δcij = | Xij−Qc5 | (56)
φcij = 101 (binary number) → δcij = | Xij−Qc6 | (57)
φcij = 110 (binary number) → δcij = | Xij−Qc7 | (58)
φcij = 111 (binary number) → δcij = | Xij−Qc8 | (59)
Δc = Σδcij (1 ≦ i ≦ 4, 1 ≦ j ≦ 4) (60)

  The quantization units 4a, 4b, and 4c confirm whether or not the processing has been completed for all the pixels in the block (step ST9). If the processing has not been completed, the processing returns to step ST6 and the processing is completed. If so, the process proceeds to step ST10.

  When the quantizing units 7a, 7b, and 7c assign the quantized data φaij, φbij, and φcij to all the pixels in the block, the encoded data creating unit 9 includes the quantized data φaij, φbij, and φcij, The quantized data with the smallest data amount is selected under the condition that the integrated values Δa, Δb, Δc of the absolute values of the errors δaij, δbij, δcij do not exceed the allowable threshold value ΔT (step ST10).

A specific example of selecting the quantized data is as follows.
Δa ≦ ΔT → quantized data φaij
Δa> ΔT, Δb ≦ ΔT → quantized data φbij
Δa> ΔT, Δb> ΔT → quantized data φcij
As described above, the quantized data having the smallest data amount is selected from among the quantized data φnij in which the integrated value Δn of the absolute value of the error is equal to or less than the allowable threshold value ΔT. Is exceeded, the quantized data φnij having the largest data amount is selected in order to preserve the image quality as much as possible.

  The quantization units 4a, 4b, and 4c check whether the processing has been completed for all the blocks (step ST11). If the processing has not been completed, the process returns to step ST2, and if the processing has been completed. The process proceeds to step ST12.

For each block, the encoded data creating unit 9 selects the selected quantized data φnij, a flag indicating the data quantization method n related to the quantized data φnij, and the pixel reference level LAn calculated by the encoding preparation unit 5n. And the encoded data is created by combining the pixel value variation width LDn (step ST12).
For example, when the quantized data φaij calculated by the quantizing unit 4a is selected, the quantized data φaij, a flag indicating the data quantization method a used by the quantizing unit 4a, and an encoding preparation unit The encoded data is created by combining the pixel reference level LAa and the pixel value fluctuation range LDa calculated by 5a.

Here, FIG. 4 is an explanatory diagram showing a configuration example of encoded data for each block generated by the encoded data generating unit 9.
A 2-bit flag indicating the used data quantization method is added to the head of the encoded data.
For example, “00” is added for the data quantization method a, “01” is added for the data quantization method b, and “10” is added for the data quantization method c.
Next, the 8-bit pixel reference level LAn calculated by the encoding preparation unit 5n and the 8-bit pixel value variation width LDn are added.

Next, when the quantized data φaij assigned by the quantizing unit 4a is selected, 1-bit × 16 quantized data φaij is added to generate encoded data.
When the quantized data φbij assigned by the quantizing unit 4b is selected, 2 bits × 16 quantized data φbij is added to generate encoded data.
When the quantized data φcij allocated by the quantizing unit 4c is selected, 3 bits × 16 quantized data φcij is added to generate encoded data.

Although FIG. 4 shows an example in which the flag indicating the data quantization method is 2-bit information, the number of bits may be determined according to the number of data quantization methods prepared in advance.
For example, if the number of data quantization methods prepared in advance is two, it is determined as 1 bit, and if it is eight, it is determined as 3 bits.
The pixel reference level LAn and the pixel value variation width LDn may be reduced to 7 bits or 6 bits or less by deleting lower bits for the purpose of reducing the code amount.

In the first embodiment, the case where the number of gradations of the pixel value level of each pixel is 2, 4, and 8 gradations (values of powers of 2) is shown. This is because a bit is assigned to each of them so that the processing can be simplified.
Therefore, when the purpose is to improve the coding efficiency without considering simplification of processing, the number of gradations assigned to each pixel may be 3 gradations or 5 gradations. In this case, for example, in the case of 3 gradation data, a bit width capable of expressing 3 to the 16th power is prepared for the pixel value data, and once the data is expressed in the form of ternary 16 digits The value is converted into a binary number and recorded.

As is apparent from the above, according to the first embodiment, the maximum value Lmax and the minimum value detected by the signal level range detection unit 2 as corresponding to a plurality of data quantization methods n prepared in advance. The pixel reference level LAn calculated by the encoding preparation units 5a, 5b, and 5c and the encoding preparation units 5a, 5b, and 5c for calculating the pixel reference level LAn and the pixel value variation width LDn of the gradation level in each block from Lmin. And the gradation threshold quantization thresholds La1, Lb1 to Lb3, Lc1 to Lc7 in each block are calculated from the pixel value variation width LDn, and a unique code is decoded from the pixel reference level LAn and the pixel value variation width LDn. Quantization threshold value calculation units 6a, 6 for calculating the gradation levels Qa1-Qa2, Qb1-Qb4, Qc1-Qc8 of each pixel , 6c and the gradation level Xij of each pixel in each block and the quantization threshold values La1, Lb1 to Lb3, Lc1 to Lc7 calculated by the quantization threshold value calculation units 6a, 6b, and 6c are compared. Quantization processing units 7a, 7b, and 7c that identify a unique code corresponding to the key level and assign the unique code to each pixel as quantized data φaij, φbij, and φcij, and quantization threshold calculation units 6a, 6b, Before the quantization data is assigned by the gradation levels Qa1 to Qa2, Qb1 to Qb4, Qc1 to Qc8 and the quantization processing units 7a, 7b, and 7c when the unique code calculated by 6c is decoded Quantization error integrating units 8a, 8b, and 8c that integrate the absolute values of errors δaij, δbij, and δcij with respect to the gradation level Xij of each pixel for each block. Of the quantized data φaij, φbij, φcij assigned by the quantization processing units 7a, 7b, 7c for each block, the data creating unit 9 calculates the error accumulated by the quantization error accumulating units 8a, 8b, 8c. Under the condition that the absolute value does not exceed the allowable threshold ΔT, the quantized data φnij having the smallest data amount is selected, the quantized data φnij, the flag indicating the quantization method n related to the quantized data φnij, and the pixel reference Since the encoded data composed of the level LAn and the pixel value variation width LDn is created, there is an effect that the encoding efficiency can be improved and the image quality of the entire image can be improved.
That is, it is possible to perform an efficient encoding process according to the characteristics of the image, and the error of the code amount is made uniform between the blocks, so that a homogeneous image with little difference in image quality can be obtained over the entire image. .

Embodiment 2. FIG.
In the first embodiment, the encoded data creating unit 9 uses the fixed allowable threshold ΔT. However, the encoded data creating unit 9 uses the target code amount per image frame and the processed code amount. The target code amount per block of the unprocessed block is calculated from the integrated code amount of the block and the number of unprocessed blocks, and the allowable threshold ΔT is determined from the target code amount per block of the unprocessed block. You may make it do.
In the second embodiment, a description will be given of a method for keeping the encoded data amount per frame constant while changing the data amount for each block according to the characteristics of the image by dynamically changing the allowable threshold ΔT. To do.

First, the encoded data creation unit 9 sets the target code amount per image frame as S, the integrated code amount of processed blocks as D, and the number of unprocessed blocks as B, according to the following equation (61). As shown, a target code amount G per block for an unprocessed block is calculated.
G = (SD) / B (61)

When the encoded data creation unit 9 calculates the target code amount G per unprocessed block, for example, a lookup table T (n) in which the correspondence between the target code amount G and the allowable threshold ΔT is recorded. ), The allowable threshold value ΔT is determined from the target code amount G.
ΔT = T (G) (62)

However, when determining the value of the lookup table T (n), the following condition is satisfied.
(1) The value of T (n) is made monotonic as the value of n increases.
That is, if the target code amount G per block is large, it is possible to increase the code amount allocated to the block. Therefore, the value of the permissible threshold ΔT is lowered, the image quality is improved, and the data amount per block is increased. Increase the control.
Conversely, if the target code amount G per block is small, control is performed to suppress the data amount per block instead of decreasing the image quality by increasing the value of the allowable threshold value ΔT.
(2) Among data quantization methods in which the value of n can be selected, a value indicating invalidity of the allowable threshold ΔT (for example, “−1”) when the amount of encoded data per block is not more than the minimum value Set.
Thereby, it is possible to prevent the code amount of one frame of the image from exceeding the target code amount S, and to always make the code amount constant.

FIG. 5 is an explanatory diagram showing the relationship between the accumulated code amount for each processed block and the encoding error allowable threshold.
In step ST12 of FIG. 2, the encoded data creation unit 9 creates final encoded data. If the accumulated code amount is less than the target code amount S per image frame, A process of adding dummy data to make the code amount of one frame of the image constant is performed.

  In the second embodiment, since the process of making the code amount per frame of the image constant is introduced, the image coding efficiency is improved by changing the data size for each block, and the original FBTC compression method. It is possible to achieve both of the advantages that the amount of encoded data is fixed.

Embodiment 3 FIG.
In the first and second embodiments, the coded data creating unit 9 out of the quantized data φaij, φbij, and φcij assigned by the quantization processing units 7a, 7b, and 7c for each block is the quantization error integrating unit 8a. , 8b, and 8c are shown for selecting the quantized data φnij having the smallest data amount under the condition that the absolute value of the error accumulated by 8b and 8c does not exceed the allowable threshold value ΔT. , The quantization data φnij assigned by the quantization processing unit 7n corresponding to a predetermined data quantization method is selected regardless of the absolute value of the error accumulated by the quantization error accumulation units 8a, 8b, and 8c. You may do it.

Specifically, the encoded data creation unit 9 creates in advance a reference table that records the correspondence between the coordinate number of a specific block for fixing the data quantization method and the data quantization method.
When acquiring the quantized data φnij for the block of the corresponding coordinate number, the encoded data creating unit 9 does not perform the process of selecting the quantized data φnij based on the allowable threshold ΔT as in the first embodiment. The data quantization method recorded in the reference table is confirmed, and the quantized data φnij assigned by the quantization processing unit 7n using the data quantization method is selected.

When the code amount per image frame is made constant as in the second embodiment, the number of unprocessed blocks is calculated when calculating the target code amount G per block of unprocessed blocks. The number of blocks for fixing the data quantization method included in the unprocessed block is subtracted from B, and the total code amount D of the processed block is included in the unprocessed block. The amount of encoded data of a block that fixes the data quantization method is added.
As a result, it is possible to remove the influence of the block whose data quantization method is fixed and dynamically change the value of the allowable threshold value ΔT.

  In the third embodiment, quantized data φnij assigned by a fixed data quantization method is obtained for a specific block designated from the outside. For this reason, the data quantization method is fixed by designating an area in which the image properties are known in advance, and stable processing independent of the image characteristics of surrounding blocks and the code amount per frame is realized. Is possible.

Embodiment 4 FIG.
FIG. 6 is a block diagram showing an image decoding apparatus according to Embodiment 4 of the present invention. In FIG. 6, the encoded data dividing unit 21 outputs encoded data (data quantization) output from the image encoding apparatus of FIG. A process of dividing the flag indicating the method n, the pixel reference level LAn, the pixel value variation width LDn, and the quantized data φnij) into M × N (M and N are natural numbers) blocks is performed. The encoded data dividing unit 21 constitutes encoded data dividing means.
The decoding unit 22 performs a process of generating image data by decoding the encoded data divided by the encoded data dividing unit 21. The decoding unit 22 constitutes a decoding unit.

The inverse quantization unit 23a performs a process of assigning a gradation level Xij corresponding to the quantized data φaij to each pixel by a data decoding method a (decoding method corresponding to the data quantization method a) prepared in advance. To do.
The inverse quantization unit 23b performs a process of assigning a gradation level Xij corresponding to the quantized data φbij to each pixel by a data decoding method b (decoding method corresponding to the data quantization method b) prepared in advance. To do.
The inverse quantization unit 23c performs a process of assigning a gradation level Xij corresponding to the quantized data φcij to each pixel by a data decoding method c (decoding method corresponding to the data quantization method c) prepared in advance. To do.

The gradation level calculation unit 24a of the inverse quantization unit 23a indicates the data quantization method a in which the flag included in the encoded data divided by the encoded data dividing unit 21 corresponds to the data decoding method a. In this case, from the pixel reference level LAa and the pixel value fluctuation width LDa included in the encoded data, the gradation levels Qa1 to Qa2 corresponding to the quantized data φaij (unique code) included in the encoded data The process of calculating is performed.
The gradation level assigning unit 25a uses the gradation level corresponding to the value of the quantized data φaij included in the encoded data among the gradation levels Qa1 to Qa2 calculated by the gradation level calculating unit 24a as Xij. A process of assigning to pixels is performed.

The gradation level calculation unit 24b of the inverse quantization unit 23b indicates the data quantization method b in which the flag included in the encoded data divided by the encoded data dividing unit 21 corresponds to the data decoding method b. In this case, the gradation levels Qb1 to Qb4 corresponding to the quantized data φbij (unique code) included in the encoded data from the pixel reference level LAb and the pixel value fluctuation width LDb included in the encoded data. The process of calculating is performed.
The gradation level assigning unit 25b uses the gradation level corresponding to the value of the quantized data φbij included in the encoded data among the gradation levels Qb1 to Qb4 calculated by the gradation level calculation unit 24b as Xij. A process of assigning to pixels is performed.

The gradation level calculation unit 24c of the inverse quantization unit 23c indicates the data quantization method c in which the flag included in the encoded data divided by the encoded data dividing unit 21 corresponds to the data decoding method c. In this case, the gradation levels Qc1 to Qc8 corresponding to the quantized data φcij (unique code) included in the encoded data from the pixel reference level LAc and the pixel value fluctuation width LDc included in the encoded data. The process of calculating is performed.
The gradation level assigning unit 25c uses the gradation level corresponding to the value of the quantized data φcij included in the encoded data among the gradation levels Qc1 to Qc8 calculated by the gradation level calculation unit 24c as Xij. A process of assigning to pixels is performed.
Note that the gradation level calculation units 24a, 24b, and 24c constitute gradation level calculation means, and the gradation level assignment units 25a, 25b, and 25c constitute gradation level assignment means.

The image data creation unit 26 selects an inverse quantization unit 23n corresponding to the flag included in the encoded data divided by the encoded data division unit 21 among the inverse quantization units 23a, 23b, and 23c, and A process of creating image data from the gradation level Xij of each pixel assigned by the gradation level assignment unit 25n of the inverse quantization unit 23n is performed. The image data creation unit 26 constitutes image data creation means.
FIG. 7 is a flowchart showing the processing contents of the image decoding apparatus according to Embodiment 4 of the present invention.

Next, the operation will be described.
When the encoded data dividing unit 21 receives the encoded data (the flag indicating the data quantization method n, the pixel reference level LAn, the pixel value variation width LDn, the quantized data φnij) from the image encoding device of FIG. The encoded data is divided into M × N blocks (M and N are natural numbers) (step ST21).
In other words, the encoded data dividing unit 21 divides the image data before being encoded by the image encoding device of FIG. 1 into corresponding portions of the divided blocks.

At this time, unlike the conventional FBTC encoding method, the encoding size of each block in the image data differs depending on the data quantization method selected for each block and is not a fixed length.
However, if the selected data quantization method is known, the encoded data size of the block corresponding to the data quantization method is uniquely determined. Therefore, the encoded data dividing unit 21 reads a flag indicating the data quantization method for each block, reads the encoded data of the corresponding data size, divides it into encoded data for one block, and then sets the next data Considering the beginning of the encoded data of the next block, the process of dividing the encoded data is performed by repeatedly performing an operation of reading a flag indicating a data quantization method for each block.

When the encoded data dividing unit 21 divides the encoded data into M × N blocks, the inverse quantizing units 23a, 23b, and 23c of the decoding unit 22 select unprocessed blocks that have not yet been decoded. One is selected (step ST22).
If the flag included in the encoded data of the selected block indicates the data quantization method a corresponding to the data decoding method a (step ST23), the gradation level calculation unit 24a of the inverse quantization unit 23a ), And input the encoded data.
That is, referring to the first 2 bits of the encoded data, if the 2-bit value is “00”, the encoded data is input.

If the flag included in the encoded data of the selected block indicates the data quantization method b corresponding to the data decoding method b, the gradation level calculation unit 24b of the inverse quantization unit 23b (step ST23). ), And input the encoded data.
That is, referring to the first 2 bits of the encoded data, if the 2-bit value is “01”, the encoded data is input.
If the flag included in the encoded data of the selected block indicates the data quantization method c corresponding to the data decoding method c, the gradation level calculation unit 24c of the inverse quantization unit 23c (step ST23). ), And input the encoded data.
That is, referring to the first 2 bits of the encoded data, if the 2-bit value is “10”, the encoded data is input.
In the fourth embodiment, since there are three types of data decoding methods that can be selected, if the 2-bit value is “11”, a decoding error is assumed.

When the gradation level calculation unit 24a of the inverse quantization unit 23a receives the encoded data of the block, the gradation level calculation unit 24a converts the pixel reference level LAa and the pixel value variation width LDa included in the encoded data into the encoded data. The gradation levels Qa1 to Qa2 corresponding to the included quantized data φaij (unique code) are calculated (step ST24a).
When the gradation level calculation unit 24b of the inverse quantization unit 23b receives the encoded data of the block, the gradation level calculation unit 24b converts the pixel reference level LAb and the pixel value variation width LDb included in the encoded data into the encoded data. The gradation levels Qb1 to Qb4 corresponding to the included quantized data φbij (unique code) are calculated (step ST24b).
When the gradation level calculation unit 24c of the inverse quantization unit 23c receives the encoded data of the block, the gradation level calculation unit 24c converts the pixel reference level LAc and the pixel value variation width LDc included in the encoded data into the encoded data. The gradation levels Qc1 to Qc8 corresponding to the included quantized data φcij (unique code) are calculated (step ST24c).

Expressions for calculating the gradation levels by the gradation level calculation units 24a, 24b, and 24c are, for example, the following expressions (63) to (76).
[In case of data decoding method a]
Qa1 = LAa−LDa / 2 (63)
Qa2 = LAa + LDa / 2 (64)
[In the case of data decoding method b]
Qb1 = LAb−LDb / 2 (65)
Qb2 = LAb−LDb / 8 (66)
Qb3 = LAb + LDb / 8 (67)
Qb4 = LAb + LDb / 2 (68)
[In case of data decoding method c]
Qc1 = LAc−LDc / 2 (69)
Qc2 = LAc−LDc × 5/16 (70)
Qc3 = LAc−LDc × 3/16 (71)
Qc4 = LAc-LDc / 16 (72)
Qc5 = LAc + LDc / 16 (73)
Qc6 = LAc + LDc × 3/16 (74)
Qc7 = LAc + LDc × 5/16 (75)
Qc8 = LAc + LDc / 2 (76)

Expressions (63) to (76) correspond to the case where the quantization threshold is calculated by Expressions (5) to (15) in the first embodiment.
Therefore, when the quantization threshold is calculated by a calculation formula other than the formulas (5) to (15), it is necessary to change the gradation level calculation formula to correspond to the calculation formula.

When the gradation level calculation unit 24a calculates the gradation levels Qa1 to Qa2, the gradation level assignment unit 25a of the inverse quantization unit 23a selects one unprocessed pixel in the block (step ST25).
Then, the gradation level assignment unit 25a selects a gradation level corresponding to the value of the quantized data φaij included in the encoded data among the gradation levels Qa1 to Qa2 calculated by the gradation level calculation unit 24a. A process of assigning to the pixel is performed as Xij (step ST26).

The gradation level assignment unit 25b of the inverse quantization unit 23b selects one unprocessed pixel in the block when the gradation level calculation unit 24b calculates the gradation levels Qb1 to Qb4 (step ST25).
Then, the gradation level assignment unit 25b selects a gradation level corresponding to the value of the quantization data φbij included in the encoded data among the gradation levels Qb1 to Qb4 calculated by the gradation level calculation unit 24b. A process of assigning to the pixel is performed as Xij (step ST26).

When the gradation level calculation unit 24c calculates the gradation levels Qc1 to Qc8, the gradation level assignment unit 25c of the inverse quantization unit 23c selects one unprocessed pixel in the block (step ST25).
The gradation level assigning unit 25c then selects a gradation level corresponding to the value of the quantized data φcij included in the encoded data among the gradation levels Qc1 to Qc8 calculated by the gradation level calculating unit 24c. A process of assigning to the pixel is performed as Xij (step ST26).

An example of a specific gradation level assignment method by the gradation level assignment units 25a, 25b, and 25c is as follows.
[In case of data decoding method a]
φaij = 0 (binary number) → Xij = Qa1 (77)
φaij = 1 (binary number) → Xij = Qa2 (78)
[In the case of data decoding method b]
φbij = 00 (binary number) → Xij = Qb1 (79)
φbij = 01 (binary number) → Xij = Qb2 (80)
φbij = 10 (binary number) → Xij = Qb3 (81)
φbij = 11 (binary number) → Xij = Qb4 (82)
[In case of data decoding method c]
φcij = 000 (binary number) → Xij = Qc1 (83)
φcij = 001 (binary number) → Xij = Qc2 (84)
φcij = 010 (binary number) → Xij = Qc3 (85)
φcij = 011 (binary number) → Xij = Qc4 (86)
φcij = 100 (binary number) → Xij = Qc5 (87)
φcij = 101 (binary number) → Xij = Qc6 (88)
φcij = 110 (binary number) → Xij = Qc7 (89)
φcij = 111 (binary number) → Xij = Qc8 (90)

The inverse quantization units 23a, 23b, and 23c confirm whether or not the processing has been completed for all the pixels in the block (step ST27). If the processing has not been completed, the processing returns to the processing of step ST25 and the processing is completed. If completed, the process proceeds to step ST28.
The inverse quantization units 23a, 23b, and 23c confirm whether or not the processing has been completed for all the blocks (step ST28). If the processing has not been completed, the processing returns to step ST22 and the processing has been completed. If so, the process proceeds to step ST29.

If the flag included in the encoded data divided by the encoded data dividing unit 21 indicates the data quantization method a corresponding to the data decoding method a, the image data creating unit 26 performs an inverse quantization unit. Image data is created by grouping the tone levels Xij of the pixels assigned by the tone level assigning unit 25a of 23a (step ST29).
When the flag included in the encoded data divided by the encoded data dividing unit 21 indicates the data quantization method b corresponding to the data decoding method b, the image data creating unit 26 Image data is created by grouping the tone levels Xij of the pixels assigned by the tone level assigning unit 25b of 23b (step ST29).
If the flag included in the encoded data divided by the encoded data dividing unit 21 indicates the data quantization method c corresponding to the data decoding method c, the image data creating unit 26 performs an inverse quantization unit. The gradation level Xij of each pixel assigned by the gradation level assignment unit 25c of 23c is collected to create image data (step ST29).

  As apparent from the above, according to the fourth embodiment, the gradation level assigning units 25a, 25b, and 25c are referred to by referring to the flags included in the encoded data divided by the encoded data dividing unit 21. Since the image data is created by combining the gradation levels Xij of each pixel assigned to any of the above, even if encoded data having a different data quantization method is created for each block, decoding is correctly performed. This makes it possible to generate a uniform image with little difference in image quality over the entire image.

Embodiment 5 FIG.
FIG. 8 is a block diagram showing an image decoding apparatus according to Embodiment 5 of the present invention. In the figure, the same reference numerals as those in FIG.
The inverse quantization unit 31 performs a process of assigning the gradation level Xij corresponding to the quantized data φnij to each pixel by any of the data decoding methods a, b, and c prepared in advance.

The gradation level calculation unit 32 is a gradation level calculation method corresponding to the flag included in the encoded data divided by the encoded data dividing unit 21, and the pixel reference level LAn included in the encoded data Processing for calculating gradation levels Qa1 to Qa2 (or Qb1 to Qb4, Qc1 to Qc8) corresponding to the quantized data φnij (unique code) included in the encoded data from the pixel value variation width LDn. carry out. The gradation level calculation unit 32 constitutes a gradation level calculation unit.
The gradation level assigning unit 33 is a gradation level assigning method corresponding to the flag included in the encoded data divided by the encoded data dividing unit 21, and the gradation level Qa1 calculated by the gradation level calculating unit 31 is used. A process of assigning to each pixel the gradation level corresponding to the value of the quantized data φnij included in the encoded data among ˜Qa2 (or Qb1 to Qb4, Qc1 to Qc8) as Xij is performed. The gradation level assigning unit 33 constitutes a gradation level assigning unit.

  In the fourth embodiment, the inverse quantization unit 23a using the data decoding method a, the inverse quantization unit 23b using the data decoding method b, and the inverse quantization unit 23c using the data decoding method c. As shown in FIG. 8, an inverse quantization unit 31 that can use any of the data decoding methods a, b, and c is provided, and the inverse quantization unit 31 performs encoding. The data decoding method to be used may be selected with reference to the flag included in the encoded data divided by the data dividing unit 21.

It is a block diagram which shows the image coding apparatus by Embodiment 1 of this invention. It is a flowchart which shows the processing content of the image coding apparatus by Embodiment 1 of this invention. It is explanatory drawing which shows an example of the value of the coefficient K1n and K2n. It is explanatory drawing which shows the structural example of the coding data for every block produced by the coding data preparation part. It is explanatory drawing which shows the relationship between the accumulated code amount for every processed block, and an encoding error allowable threshold value. It is a block diagram which shows the image decoding apparatus by Embodiment 4 of this invention. It is a flowchart which shows the processing content of the image decoding apparatus by Embodiment 4 of this invention. It is a block diagram which shows the image decoding apparatus by Embodiment 5 of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Image data division part (image data division means), 2 Signal level range detection part (Signal level range detection means), 3 Encoding part (encoding means), 4a, 4b, 4c Quantization part, 5a, 5b, 5c Encoding preparation unit (encoding preparation unit), 6a, 6b, 6c Quantization threshold calculation unit (quantization threshold calculation unit), 7a, 7b, 7c Quantization processing unit (quantization unit), 8a, 8b, 8c Quantum Encoding error integrating unit (quantized error integrating unit), 9 Encoded data generating unit (encoded data generating unit), 21 Encoded data dividing unit (encoded data dividing unit), 22 Decoding unit (decoding unit), 23a, 23b, 23c Inverse quantization unit, 24a, 24b, 24c Gradation level calculation unit (gradation level calculation unit), 25a, 25b, 25c Gradation level allocation unit (gradation level allocation unit), 26 Image data creation (Image data generating means), 31 inverse quantization unit 32 gradation-level calculation section (gradation level calculating means), 33 gray level assignment unit (gradation level assignment means).

Claims (2)

The gradation level in each block with reference to the gradation level of each pixel indicated by the image data dividing means for dividing the image data output from the image sensor into blocks and the image data divided by the image data dividing means Using the signal level range detection means for detecting the maximum value and the minimum value of the signal, and the maximum value and the minimum value detected by the signal level range detection means, the gradation level of each pixel in each block is quantized and encoded. In the image coding apparatus comprising coding means for creating data, the coding means is detected by the signal level range detection means as corresponding to a plurality of data quantization methods prepared in advance. A plurality of encoding preparation means for calculating the reference level and fluctuation range of the gradation level in each block from the maximum value and the minimum value, and the above encoding preparation The gradation level quantization threshold in each block is calculated from the reference level and fluctuation range calculated by the stage, and the unique code assigned as quantized data is decoded from the reference level and fluctuation width. A plurality of quantization threshold calculation means for calculating the gradation level of each pixel, and comparing the gradation level of each pixel in each block with the quantization threshold calculated by the quantization threshold calculation means, A unique code corresponding to the key level is identified, and the unique code calculated by the quantization threshold value calculating means is decoded by a plurality of quantization means for assigning the unique code to each pixel as quantized data. The absolute value of the error between the gradation level of each pixel and the gradation level of each pixel before the quantization data is assigned by the quantization means is integrated for each block. And an absolute value of the error accumulated by the quantization error accumulation unit among the quantized data allocated by the plurality of quantization units for each block. Calculated by the quantized data, the method identification information indicating the quantization method related to the quantized data, and the encoding preparation means, by selecting the quantized data having the smallest data amount under the condition not exceeding the threshold. Encoded data generating means for generating encoded data having a reference level and a fluctuation range, wherein the encoded data generating means includes a target code amount per image frame, an integrated code amount of processed blocks, The target code amount per block of the unprocessed block is calculated from the number of unprocessed blocks, and allowed from the target code amount per block of the unprocessed block. An image encoding apparatus characterized by determining a threshold value. The gradation level in each block with reference to the gradation level of each pixel indicated by the image data dividing means for dividing the image data output from the image sensor into blocks and the image data divided by the image data dividing means Using the signal level range detection means for detecting the maximum value and the minimum value of the signal, and the maximum value and the minimum value detected by the signal level range detection means, the gradation level of each pixel in each block is quantized and encoded. In the image coding apparatus comprising coding means for creating data, the coding means is detected by the signal level range detection means as corresponding to a plurality of data quantization methods prepared in advance. A plurality of encoding preparation means for calculating the reference level and fluctuation range of the gradation level in each block from the maximum value and the minimum value, and the above encoding preparation The gradation level quantization threshold in each block is calculated from the reference level and fluctuation range calculated by the stage, and the unique code assigned as quantized data is decoded from the reference level and fluctuation width. A plurality of quantization threshold calculation means for calculating the gradation level of each pixel, and comparing the gradation level of each pixel in each block with the quantization threshold calculated by the quantization threshold calculation means, A unique code corresponding to the key level is identified, and the unique code calculated by the quantization threshold value calculating means is decoded by a plurality of quantization means for assigning the unique code to each pixel as quantized data. The absolute value of the error between the gradation level of each pixel and the gradation level of each pixel before the quantization data is assigned by the quantization means is integrated for each block. And an absolute value of the error accumulated by the quantization error accumulation unit among the quantized data allocated by the plurality of quantization units for each block. Calculated by the quantized data, the method identification information indicating the quantization method related to the quantized data, and the encoding preparation means, by selecting the quantized data having the smallest data amount under the condition not exceeding the threshold. reference level and includes a coded data generating means for generating encoded data comprising a variation width, the coded data generating means, for certain block designated from the outside, is integrated by the quantization error integrating means and regardless of the absolute value of the error, to and selects the quantized data allocated by the quantizing means corresponds to the predetermined data quantization method That picture coding apparatus.

Images