JPH0771292B2 - Image adaptive quantization coding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Outline] A high-efficiency adaptive quantization coding method for processing an image signal in block units, and a scalar quantization circuit which inputs an input signal (x _i ) and operates in parallel. , A vector quantizer, and a plurality of quantizers, each of which is a combination of the vector quantizer and the scalar quantizer in cascade connection, and a plurality of quantizers each having a quantizer output from each quantizer as an input A generated information amount calculation circuit for calculating the generated information amount of the circuit, and a quantization distortion calculation circuit for calculating the quantization distortion of each quantization circuit with a plurality of each quantized output and input signal as input, and the generated information amount calculation An evaluation / selection circuit, which receives as input the amount of generated information for each quantizer output from the circuit and the value of the quantization distortion for each quantizer output from the quantization distortion calculation circuit, is provided. The selection circuit is set The output selected from the plurality of quantization circuits is adaptively controlled according to the specified standard, and the standard is set such that the quantization distortion is equal to or less than the specified threshold and the generated information amount is minimum.

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a highly efficient adaptive quantization coding method when processing an image signal in block units.

A high-efficiency coding method for transmitting an image signal at a low bit rate is required for effective use of a transmission line and reduction of communication cost. It is naturally desired to perform high-efficiency encoding even when moving images are transmitted by the interframe encoding method.

On the other hand, in recent years, various methods have been used as the image signal quantization method, but the characteristics of the image signal, the characteristics of the transmission path, the usage form, etc. are not always fixed. It is necessary to always realize highly efficient quantization coding corresponding to.

[Prior Art] Conventionally, as a method for quantizing an image signal, a scalar quantizing method (SQ) for quantizing pixel by pixel, or a pattern with the smallest distortion among patterns prepared in advance by dividing pixels into blocks. There was a vector quantization method (VQ) etc. which encodes the index of.

In general, the scalar quantization method (SQ) has a large amount of information but little distortion, and the vector quantization method (VQ) tends to have a small amount of information but large distortion, which varies depending on the nature of the input signal.

[Problems to be Solved by the Invention] As described above, according to the conventional encoding method, both the scalar quantization method (SQ) and the vector quantization method (VQ) have advantages and disadvantages, and the nature of the input signal. There is a problem in that there is no coding method having a small amount of information and a small distortion and having a high coding efficiency.

It is an object of the present invention to provide an adaptive quantization coding method for an image that can switch to a most suitable quantization code according to an input signal when blocking and quantizing a moving image signal or its prediction error signal. To aim.

[Means for Solving Problems] FIG. 1 is a diagram showing a principle configuration of the present invention.

In the figure, 11 is a scalar quantizer (SQ1), 12 is a vector quantizer (VQ2), 13 is a quantizer that combines a vector quantizer (VQ3) and a scalar quantizer (SQ3), and 1n is a vector. A combination quantizer different from the above 13 such as an encoder configured by connecting only quantizers in multiple stages or an encoder configured by a plurality of scalar quantizers having different quantization characteristics (n is an arbitrary number) Number), 16 is a generated information amount calculation circuit, 17 is a quantization distortion calculation circuit, 18 is an evaluation / selection circuit, and 19 is a switching circuit.

The present invention provides a plurality of quantization circuits (11 to 1n) each having a different encoding method or quantization characteristic as a circuit for quantizing an input image signal, and operates these in parallel or in time division. The amount of information and the quantization distortion are calculated for each coded output, and each quantizing circuit is evaluated, and the optimum quantizing circuit that satisfies the evaluation criterion at each time point is selected and output.

[Operation] In FIG. 1, the case where an image signal or its prediction error signal is supplied as the input signal x _i (100) and the image signal is now input will be described.

The input signal x _i has a plurality of quantizers 11 (SQ1), 12 (VQ2),
13 (VQ3 + SQ3), ... is input in parallel to 1n, the quantized output of the quantizer 11 is q _1i , and the quantized output of the quantizer 12 is
q _2i , the quantizer 13 outputs SQ when the quantized output of VQ 3 is q _31i .
The quantized output of 3 is q _32i and the quantized output of quantizer 1n is q _ni
Is. Note that the output of VQ2 and VQ3 of the vector quantizer q _2i
And q _31i represent both the index of each output vector and its pixel pattern, and each index of SQ and VQ is assigned a variable length code such as Huffman code.

Outputs of multiple quantizer sets, q _1i , q _2i , q _31i + q _32i ,
・ ・ ・ Path 1, path 2, path 3 respectively on the q _ni signal paths
..Pass 1 in the generated information amount calculation circuit 16
The amount of information I _{1 to} I _n when each path n passes is calculated. Also, performing calculations by the quantization distortion _{d 1, d 2, d 3} ··· d n , for example following expression in the quantization distortion calculation circuit 17 in each pulse.

(However, i represents the block number of the image signal.) Thus, the generated information amount calculation circuit 16 and the quantization distortion calculation circuit
The information amount I _N , which is the result calculated in 17, and the quantization distortion d _N (where N represents the path number, in this case N = 1 to n),
The optimum path N is sent to the evaluation / selection circuit and evaluated according to the evaluation standard at that time.

When any path is selected in the evaluation / selection circuit 18, a quantization selection signal 180 indicating which path is selected is displayed.
Is generated and transmitted, and is also supplied to the switching circuit 19 to switch and transmit the quantized signal of the path selected from among the paths of the input quantizers as the output signal 190.

Evaluation condition as a reference of evaluation is given to the evaluation and selection circuit 18 from the evaluation criteria set input 181, as a condition, and specifying a certain amount of information D, among the path N to be the I _N <D, the I _N A method of selecting a path N that minimizes, determines a certain distortion value H, and selects a path n that minimizes d _N among paths N that satisfies d _N <H.
Method of selecting, monitor the buffer capacity of the transmission line,
There is a method of controlling the amount of generated information by changing the evaluation function according to the value.

[Embodiment] FIG. 2 shows the configuration of a motion-compensated interframe coding method for a moving image, which is an embodiment of the adaptive quantization coding method for an image according to the present invention.

In this embodiment, three quantizers are provided, which are a scalar quantizer, a vector quantizer, and a quantizer combining a vector quantizer and a scalar quantizer.

In FIG. 2, 20 is an input image signal, 21 is a subtraction circuit, 22
Is a constituent element in the adaptive quantization coding unit,
Is a scalar quantizer (SQ1), 222 is a vector quantizer (VQ2), 223 is a vector quantizer (VQ3), 224 is a scalar quantizer (SQ3), 225 is a subtractor, and 226 to 229 are the first It corresponds to each circuit of reference numerals 16 to 19 in the figure, 23 is a motion detection unit between moving image frames, 24 is a variable delay circuit, 25 is a frame memory, 26 is an addition circuit, 27 is a quantizer selection signal , 28
Is a prediction error information signal, and 29 is a motion vector information signal.

In FIG. 2, the input image signal 20 is input to the subtraction circuit 21, the difference from the predicted image signal of the previous frame from the frame memory 25 is obtained by the subtraction circuit 21, and the obtained prediction error signal is adaptively quantized and encoded. Supply to part 22. On the other hand, the motion detector 23 detects motion information based on the image signal of the previous frame from the frame memory 25 and the input image signal 20, and generates and transmits a motion vector information signal 29.

The adaptive quantization coding unit 22 is a quantizer having a plurality of quantization paths as shown in FIG. 1, and the prediction error signal which is the output of the subtraction circuit 21 input to this quantizer is It has various properties depending on the size of the movement and the input pattern.

In order to adaptively deal with these input signals, the amount of generated information of the output pulse of three sets of the scalar quantizer SQ1, the vector quantizer VQ2, and the combination of the scalar quantizer SQ3 and the vector quantizer VQ3, and Obtaining the quantization distortion in the generated information amount calculation circuit 226 and the quantization distortion calculation circuit 227, respectively,
The evaluation / selection circuit 228 evaluates according to the evaluation standard input from the evaluation standard setting input 220 (evaluation method will be described later).
Then, the optimum path is selected, and the prediction error information signal 28 is transmitted from the switching circuit 229 together with the quantizer selection signal 27. In addition,
The scalar quantizers and the vector quantizers used here are all known in the related art, and the detailed description of their specific configurations is omitted.

Next, an evaluation method of various quantization encoders in the configuration of the embodiment shown in FIG. 2 will be described using a concrete example shown in FIG.

The block size for encoding the image signal is 4 (pel: picture element / pixel) × 4 (line), and the inter-frame difference signal is X _i (i = 1 to 16).
The respective outputs of 3 are Y _1i , Y _2i , Y _3i , the power of the quantization error is d ₁ , d ₂ , d ₃ , and the generated information amount is I ₁ , I ₂ , I ₃ . Where n
Indicates the quantization path, Is.

Also, in this example, the vector quantizers VQ2 and VQ3 use 2 ¹⁰ patterns, that is, a 10-bit code,
The scalar quantizers SQ1 and SQ3 are variable length codes having a short bit length near 0 and a long bit length otherwise.

As a specific example of the inter-frame difference signal X _i which is an input to the adaptive quantization encoding unit 22, a case where the signal shown in FIG. 3A is given will be described.

With respect to the input X _i , the operation of the scalar quantizer SQ1 shown in FIG. 2 produces a coded output as shown in (A) of FIG. The number of bits for encoding (not just binary code) the numerical values in each of the 4 × 4 = 16 columns is shown in parentheses, and the total number of these bits is 46 bits. The information amount I _{1 of the} path 1 shown in (a) of FIG.

Next, for the same input X _i , the vector quantizer VQ2 of FIG.
Generates a pattern as shown in FIG. 3 (B) by the quantizing operation, but since this is displayed by any of 2 ¹⁰ (1024) patterns, an information amount of 10 bits is required. The information amount I _{2 of the} path 2 shown in (b) of C) is obtained.

When vector quantization is first performed on the same input X _i by the vector quantizer VQ3, 10-bit information is generated in the same manner as in the above-mentioned path 1, and then from the input X _i to the vector quantizer VQ.
The scalar quantizer SQ3 performs scalar quantization on the remaining signals except the components quantized in 3, and the result is a 25-bit output as shown in the lower stage of (c) of FIG. 3B, From pass 3 at this point in Fig. 3 (C)
The information amount I ₃ is 35 bits in total as shown in (c).

Thus, Path 1 requires 46 bits of information, and Path 2
It is calculated in the generated information amount calculation circuit 226 in FIG. 2 that the information amount of 10 bits is required and the information amount of the path 3 is 35 bits.

Further, when the quantization error of each path is calculated in the quantization distortion calculation circuit 227 according to the above-described calculation formula of the quantization error d _n ,
In the case of this specific example, as a result, the quantization distortion d ₁ of path 1 is calculated as 1.3, the quantization distortion d ₂ of path 2 is calculated as 10.3, and the quantization distortion d ₃ of path ₃ is calculated as 1.8. Information amount I _{1 to} I ₃
It is also shown in FIG. 3 (C).

When these calculation results are input, the evaluation / selection circuit 228
The evaluation is performed on the basis of the evaluation standard, and there are the following methods as the evaluation method adopted at that time.

Evaluation method example 1 To keep a certain frame rate (number of dropped frames) and send a moving image even if the image quality is reduced, if you want to suppress the amount of information generated per block, set a certain amount of information. Set, for example, if this is 30 (bits), I _n
Among the paths n satisfying ≦ 30, the path n having the smallest quantization distortion d _n is selected, and when there is no one satisfying I _n ≦ 30, the path n having the smallest I _n is selected. In the case of FIG. 3, according to (C), since only the path 2 in (b) corresponds to the quantized encoder having the information amount of 30 bits or less, the path 2 (VQ2) is selected (n =
1).

Evaluation Method Example 2 When it is necessary to maintain a certain level of image quality, it is required to keep the quantization distortion per block low, and therefore a certain low quantization distortion is set. For example, if this was the 3, in the path n satisfying d _n ≦ 3, select the path n the amount of information I _n is minimized, if there is no satisfying d _n ≦ 3 is, d _n is the minimum , The path n is selected.

In the case of FIG. 3, according to (C), a quantized encoder having a quantization distortion of 3 or less has d ₁ = 1.3 in (a) and d ₃ = 1.8 in (c).
Since one path satisfies the condition, the information amount I ₃ = 35 (bits) of path 3 is smaller than the information amount I ₁ = 46 (bits) of path 1 if the one with less information amount is selected. Eventually, pass 3 is selected.

Evaluation Method Example 3 A certain evaluation function f is determined, and a path n that minimizes f (I _n , d _n ) is selected. In this case, the numerical values of both the information amount I _n and the quantization distortion d _n are set to a certain constant numerical value and a path satisfying both conditions is selected.

EFFECTS OF THE INVENTION According to the present invention, when quantizing an image signal or a prediction error signal thereof, it is possible to adaptively select a quantization coding path according to an input signal, thus improving the coding efficiency of an image. You can Further, it is possible to adaptively perform highly efficient coding also in motion compensation interframe coding of a moving image.

[Brief description of drawings]

FIG. 1 is a diagram showing a principle configuration of the present invention, FIG. 2 is a diagram showing a configuration of an embodiment of the present invention, and FIG. 3 is a diagram showing an operation example of each quantizer in the embodiment. In FIG. 1, 11: Scalar quantizer 12: Vector quantizer 13: Combination quantizer 1v of vector quantizer and scalar quantizer 1n: Other combination quantizer 16: Generated information amount calculation circuit 17: Quantization distortion calculation circuit 18: Evaluation / selection circuit 19: Switching circuit

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Kiichi Matsuda 1015 Kamiodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture, Fujitsu Limited (72) Inventor Toshitaka Tsuda 1015, Kamikodanaka, Nakahara-ku, Kawasaki, Kanagawa Prefecture, Fujitsu Limited ( 56) References JP-A-58-197984 (JP, A)

Claims (2)

[Claims] 1. An adaptive quantization coding system for an image, comprising a quantization circuit for quantizing an image signal according to a plurality of different quantization systems, wherein an input signal (x _i ) is inputted to operate in parallel. A plurality of quantizer circuits (11 to 1n), which are a scalar quantizer circuit, a vector quantizer circuit, and a combination circuit of cascade connection of the vector quantizer circuit and the scalar quantizer circuit; A generated information amount calculation circuit (16) for calculating the generated information amount of each of the plurality of quantization circuits with a quantized output as an input, and each of the plurality of quantized outputs and an input signal (x _i ) as an input A quantizing distortion calculating circuit (17) for calculating the quantizing distortion of the quantizing circuit, the generated information amount for each quantizer output from the generated information amount calculating circuit (16), and the quantizing distortion calculating circuit ( 17) each quantizer output from And an evaluation / selection circuit (18) that receives the value of the quantization distortion of the input, and the evaluation / selection circuit adaptively controls an output selected from the plurality of quantization circuits according to a set designation standard. An adaptive quantization coding method for an image is characterized in that a criterion that a quantization distortion is equal to or less than a designated threshold and a generated information amount is minimum is set as the designation criterion. 2. The adaptive quantization of an image according to claim 1, wherein a criterion that an amount of generated information is equal to or less than a designated threshold and a quantization distortion is minimum is set as the designation criterion. Encoding method.