JP5201035B2 - Image encoding device - Google Patents

Description

  The present invention relates to an image coding apparatus that compresses image data by wavelet transform and entropy coding.

  JPEG2000 is an image coding method in which image data is wavelet transformed and quantized, and the quantized data is modeled in bit plane units, and then MQ coding, which is a kind of arithmetic coding, is performed. Compared with the conventional JPEG method, the coding efficiency is good, and since wavelet conversion is adopted, block noise is hardly generated.

  Furthermore, after MQ coding, a final code stream can be generated after post-quantization in which codes are discarded in units of bit planes (more precisely, sub-bit planes called paths), so that code amount control can be realized relatively easily. There is also an advantage. In general, the code amount control method is such that a rate is obtained so that an image with the best image quality within a specified code amount, that is, a code that decodes an image with the least error from the input image can be obtained.・ Strain optimization techniques are often adopted.

  However, strictly controlling the rate and distortion increases the processing time and requires a large amount of memory. In order to solve these problems, in Patent Document 1 (Japanese Patent No. 4064279), a plurality of tables describing bit planes to be discarded for each subband are prepared, and the optimum table is based on the code amount after MQ coding. A method of selecting is proposed.

  FIG. 17 includes a code amount control unit that prepares a plurality of tables describing bit planes to be discarded for each subband and selects an optimum table based on the code amount after MQ coding. A configuration example of an image encoding device is shown. In this configuration example, different types of data such as wavelet coefficient data and MQ code data are temporarily stored in the same memory 2. In this case, an arbitration circuit for arbitrating access to the memory 2 between a plurality of functional blocks is provided in the preceding stage of the memory. Naturally, it is also possible to prepare a separate memory for each data such as wavelet coefficient data and MQ code data.

  The input image data is converted into wavelet coefficient data for each subband by two-dimensional wavelet conversion, and stored in the memory 2. When the image data is that of a color image, the wavelet transform is performed independently on the luminance (Y) data and the color difference (Cb, Cr) data, and the result is shown in FIG. Each subband in a component (hereinafter referred to as a component) is stored in the memory 2.

  The coefficient data of each subband is further divided into rectangular areas called code blocks, quantized in units of this code block, and then divided into bit planes and subjected to MQ coding.

  In FIG. 18 (2), image data is converted into coefficient data for each subband by two-stage wavelet transform (division level 5) by two-dimensional wavelet transform, and each subband is a code block having a size of 32 × 32. An example of division is shown below. When the subband size is smaller than 32 × 32 such as 5LL to 3HH, these subbands are not divided and processed as one code block.

  FIG. 18 (3) is a diagram illustrating the concept of bit plane division. The coefficient data included in the code block is expressed as an absolute value, and 32 × 32 bits at the same bit position are used as a bit plane, and there is non-zero data in the bit plane, but all bits are higher than that. MQ coding is sequentially performed from the bit plane in which is 0 to the lowest bit plane. In the example of FIG. 18 (3), all data from the most significant bit to the bit plane with the bit plane number Bp + 1 is 0 (these are called zero bit planes), and the bit plane with the bit plane number Bp is the first non-zero. This is when there is data. At this time, the data from the bit plane Bp to the bit plane 1 is MQ-encoded.

Further, the bit data of one bit plane is divided into sub-bit planes called three paths depending on the influence on the image quality at the time of decoding, and encoding is performed with priority over data belonging to a path that seems to have a large influence. This situation is shown in FIG. In each bit plane, the first encoded bit is a bit belonging to a path called “significance propagation”, the second encoded is a bit belonging to a path called “magnitude refinement”, and the last encoded bit These bits belong to a path called “cleanup” that does not belong to any of the above. Since all bits included in the most significant bit plane to be encoded (bit plane Bp in FIG. 18 (4)) are included in the cleanup path, the number of paths to be encoded in one code block is
3 × (Bp−1) + 1 = 3 × Bp−2
It becomes.

  As shown in FIG. 19, the MQ code that is the result of the MQ coding is stored in the memory 2 for each code block included in each component and each subband. At the same time, the number of zero bit planes (#ZBP), the number of encoded paths (#CP), and the code amount (len) of the MQ code for each path are also stored in the memory as information necessary for encoding. Is done.

  Simultaneously with the MQ coding, data of the total code amount (Q) and the code reduction amount (Σlen) for each pass is generated for code amount control. The total code amount is the code amount of the MQ code at the time when encoding of all code blocks is completed, and the code reduction amount for each pass is the same for all code blocks included in each subband of each component. This is the sum of the code amount (len) for the pass of the pass number.

FIG. 20 (1) shows a more detailed block diagram of the encoding control / bitplane dividing unit 24, MQ encoding unit 26, and code amount calculation unit 16. After completion of the wavelet, the encoding control / bit plane dividing unit 24 reads the quantized coefficient for each component and each subband code block, divides the bit plane, and converts the bit plane data to the MQ encoding unit 26. The code amount calculation unit 16 controls the code amount calculation operation at the same time. FIG. 20 (2) shows a control flow of the encoding control / bit plane dividing unit 24. Here, C is a component number (Y, Cb, Cr), SB is a subband number (correspondence between an actual subband name and number is shown in FIG. 20 (3)), and CB is a code. Indicates the block number. The maximum value of the number of code blocks included in a subband is determined for each component (C) and subband (SB) if the image size and the wavelet division level are determined. This
Cblast (C, SB)
It expresses. In the flow of FIG. 20 (2), after starting, code block data is read and MQ coding is performed in units of code blocks in each subband in the order of Y5LL → Y5HL → Y5LH →... → Cr1HH. In reading the code block data, the wavelet coefficient data included in the code block to be processed is read from the memory 2 and written to the bit lane division memory after quantization. At the same time, the quantized coefficient data included in the code block is used. The maximum number of bit planes, that is, the number of bit planes (Bp) to be encoded with the code block is obtained. FIG. 21 shows an encoded bit plane number calculation circuit 32. Bits of the quantized coefficient data (Q) are connected to qn to q1 and input to the F / F 34. By inputting the bit-wise OR between the output of the F / F 34 and the next coefficient data to the F / F 34, the bit that becomes “1” is stored in the F / F 34 through all the code blocks, and is encoded by the priority encoder. By doing so, the bit which is “1” on the most significant side can be obtained.

  FIG. 22 shows a processing flow of MQ coding in units of code blocks. After the start, the number of coded passes (CP) is calculated from the number of coded bit planes (Bp) (S2202). If there is a bit plane to be encoded (S2204), the bit plane is read from the bit lane dividing memory, and the MQ encoding unit 26 performs bit modeling and MQ encoding as a cleanup path (S2206). When the encoding of the pass is completed, the pend indicating that the encoding of the pass is completed is set to “1” (S2208). This signal is used when the code amount calculation unit 16 calculates the code amount for each pass. If the encoded pass is the last encoding pass, the processing is terminated (S2210). Otherwise, the next bit plane is encoded. Since the bit planes other than the most significant bit are MQ-encoded in three passes of “significance propagation”, “magnitude refinement”, and “cleanup”, the flow of the encoding control unit 24 is performed by bits from the bit lane dividing memory. Plane reading and MQ encoding are performed three times. The MQ encoder 26 performs bit modeling and MQ encoding for each path. Furthermore, every time encoding in one pass is completed, the pend is set to “1” (S2216, S2222). When encoding in the cleanup pass is completed, it is confirmed whether encoding of all passes is completed (S2210). If not yet completed, encoding of the next bit plane is performed.

  FIG. 23 (1) shows an internal circuit of the code amount calculation unit 16. The valid signal is generated by the MQ encoder 26 and asserted at a timing when the MQ code data becomes valid. The code amount calculation unit 16 calculates the total code amount (Q) and the code reduction amount (Σlen) for each path by counting the valid signals. The total code amount (Q) is calculated by counting the number of valid assertions over the entire image. The code reduction amount (Σlen) for each pass is determined by counting valid signals asserted by encoding a code block, and at the time when encoding of one pass is completed (pend becomes “1”). In this case, the code reduction amount stored in the same address encoded so far is read from the memory, and the count value is added to it. Write the value. As a result, the code reduction amount data is stored in the memory for each component, each subband, and each path, as shown in FIG.

  When MQ encoding is completed, the final code stream is generated after code amount control is performed. JPEG2000 does not use all the encoded data after MQ encoding, but performs post-quantization that discards the code in units of paths in order to match a predetermined code amount, and then uses the remaining path codes to generate the final code stream. Is generated. Here, a method of code amount control using a truncation table will be described.

  FIG. 24 shows an example of the truncation table. The truncation table is a table in which the number of paths to be discarded when generating the final code stream using the code after MQ coding is described for each subband. The numbers in FIG. 24 indicate how many codes from the lower order of the code blocks belonging to the corresponding subband are discarded to generate the final code stream. For example, the table with the table number T0 indicates that the code for the lowest one pass is discarded in all code blocks included in the subband 1HH of the component Cb. Furthermore, a plurality of tables in which the number of paths to be discarded in any one subband is increased by one are prepared and arranged in order from the smallest table number. In the example of FIG. 24, the table number T2 has a larger number of paths to be discarded by the component Cb and subband 1HH by one than the table of T1.

In the rate control circuit unit 18 of FIG. 17, the final code amount is set to the set code amount (N) using the total code amount (Q) generated by the code amount calculation unit 16 and the code reduction amount (Σlen) for each pass. One table is selected from a plurality of truncation tables (T0, T1,...) So as to be the maximum within a range not exceeding. FIG. 25 shows an example of a control flow of the rate control circuit unit 18. The variables in FIG. 25 have the following contents, respectively.

  After starting, each variable is initialized (S2502), and the total code amount (Q) is compared with the set code amount (N) (S2504). When the total code amount is larger, the flow for selecting a truncation table is entered. First, the number of paths to be discarded for each component and subband that has been selected is moved to TPB (TPB ← TP), and data of table number T is newly read from the memory 2, and the number of paths to be discarded for each component and subband. Is set to TP (TP ← TBL (T)) (S2506). Next, the number of paths to be discarded differs between TP and TPB (S2508), and subbands are searched. TP and TPB always have subbands with different values, and the code reduction amount of the pass number CP of the corresponding subband is read from the memory of the code amount calculation unit 16 (S2510), and the total code amount (Q ) (S2512) and compared with the set code amount (N) (S2514). If the total code amount (Q) is larger (YES in S2514), the value of T is incremented (S2518) and the same process is repeated. If the set code amount (N) is large (NO in S2514), the final truncation path number (TP) is output to the packet circuit generation circuit, and the process ends.

  According to the truncation table selected by the rate control circuit unit 18, the final code stream is generated by the packet generation circuit 20 and the code formation unit 22.

  Here, MQ coding is the most burdensome process among the processes related to JPEG2000. Nevertheless, since the amount of code after MQ coding is not known until it is actually coded, in order to control the amount of code, as shown above, It is necessary to perform MQ encoding. As a result, there are codes that have been used up to MQ coding but have not been used in the final code stream controlled to have a predetermined code amount.

  It is possible to speed up the entire encoding process by performing MQ encoding only on the bit planes determined in advance for each subband. In this case, however, the bit plane that should be originally required depending on the image is encoded. May not be included in the final code stream and may lead to degradation of image quality.

  In the apparatus of Patent Document 2 (Patent No. 40003628), the coefficient data after frequency conversion is divided into bit plane units, and the number of data that becomes non-zero first in each bit plane is calculated to calculate the code amount. After the estimation, only the bit plane that is predicted to be encoded is encoded to improve processing efficiency. However, even if the bit plane to be encoded is determined only by the code amount estimated by this method and the code amount is controlled, there is a possibility that the optimum image quality cannot be obtained as a result.

  In addition, the following are mentioned as prior literature of this application.

  Patent Document 1 discloses a system in which a plurality of tables describing bit planes to be discarded for each subband are prepared and an optimum table is selected based on the code amount after MQ coding.

  Also, the coefficient data after frequency conversion is divided into code blocks, and the number of coefficient data whose value in the bit plane is non-zero and the value in the higher bit plane is 0 is counted for each bit plane. Then, an estimated code amount is obtained based on the value, and the estimated code amount is added until a predetermined target code amount is obtained to predict a bit plane to be encoded, and the bit plane to be encoded is arithmetically calculated. Patent Document 2 discloses an image encoding device that generates an encoded code stream from an encoded arithmetic code.

  Also, an image encoding device that calculates a bit plane to be encoded based on the number of encoded bit planes selected in a code block at the same position of the immediately preceding image in time encoding in continuous image and moving image encoding Patent Document 3 is disclosed as a disclosure of the above.

  The present invention provides an image encoding device that increases the processing speed by predicting a code that is not used in the final code stream and performing MQ encoding while minimizing the influence on image quality. Objective.

The present invention has been made to achieve the above object. An image encoding apparatus according to claim 1 of the present invention is provided.
A frequency conversion means for converting the frequency of the image data;
Bit plane encoding means for encoding the coefficient data after frequency conversion into a rectangular area called a code block for each bit plane after being divided;
Storage means for storing a plurality of tables describing combinations of the number of bit planes to be discarded with respect to the code after bit plane encoding;
In an image encoding device including code amount control means for controlling a code amount by selecting a set of tables that are equal to or less than a predetermined code amount from the plurality of tables,
Furthermore, table selection result prediction means for predicting a set of tables selected when the entire image is encoded using the code amount at the time when the encoding of the code block included in the predetermined subband is completed; ,
Coding bit plane control means is provided for controlling the coding means so as to determine bit planes to be coded with code blocks included in the remaining subbands based on the prediction result.

An image encoding apparatus according to claim 2 of the present invention is provided.
The image encoding device according to claim 1,
From the set code amount used in the code amount control and the size of the predetermined subband, a prediction code amount assigned to the corresponding subband is calculated,
From the plurality of tables describing combinations of the number of bit planes to be discarded to be used in the code amount control, using the code amount for each bit plane when the encoding of the code block included in the subband is completed, Select the table with the maximum number of discarded bitplanes in the range where the total subband code amount does not fall below the predicted code amount,
Code blocks included in the remaining subbands are encoded using the number of bitplanes discarded in the corresponding subband described in the selected table.

An image encoding apparatus according to claim 3 of the present invention is provided.
The image encoding device according to claim 2,
Using the average value of the wavelet coefficient data for each subband, the prediction code amount allocated to the coded subband is calculated.

An image encoding apparatus according to claim 4 of the present invention is provided.
The image encoding device according to claim 2,
A plurality of second tables describing a combination of the number of bit planes to be discarded only for the previously encoded subbands and data specifying the table used in the code amount control corresponding to each table. Use to predict the table selection result.

An image encoding apparatus according to claim 5 of the present invention is provided.
The image encoding device according to claim 2,
A table selection result prediction is performed by calculating the number of bit planes to be discarded using a predetermined calculation formula.

  The image encoding apparatus according to the present invention predicts a path that is expected to be discarded in the final code stream by using the information of the code amount at the time when encoding up to the middle subband is completed, and the result Since the number of passes for encoding the remaining subbands is reduced based on the above, the encoding process can be speeded up in a range that does not depend on the image and does not affect the image quality.

  In addition, since the image encoding apparatus of the present invention uses a table used for code amount control as a selection result predicting unit, it is possible to minimize the amount of increase in circuit and the amount of change of the control unit due to this function addition. it can.

  In addition, the image coding apparatus of the present invention calculates the code amount allocated to the subbands encoded in advance using the result of wavelet transform of the entire image, so that the accuracy of the prediction result is improved.

  In addition, the image coding apparatus according to the present invention performs selection result prediction using a table in which the number of passes to be discarded corresponding to only pre-encoded subbands is described, so that the time required for selection result prediction is shortened. be able to.

  Furthermore, the image encoding apparatus of the present invention calculates the number of paths to be discarded in the subband from a simple arithmetic expression given instead of using a table, so that the circuit scale can be reduced.

It is a block diagram of the image coding apparatus provided with the table selection result prediction means based on the 1st Embodiment of this invention. It is an operation | movement flow of the encoding control part in the image coding apparatus which concerns on the 1st Embodiment of this invention. Even if the paths described in C = 1, 2, 3, SB = 1 (5LL) to 10 (3HH) are discarded, the code amount does not become smaller than a certain target value, and SB = 1 (5LL) 10 (3HH) is the largest and SB = 11 (2HL) to 16 (1HH) is the smallest table, and the SB = 11 (2HL) to 16 (1HH) value of the table is selected. It is the figure which showed the relationship of setting it as a predicted value typically It is an operation flow of selection result prediction. The number of coefficient data of each subband and its average value, and further, setting a weight for each subband and multiplying them for each subband, the ratio of the sum of 5LL to 3HH and the sum of 2HL to 1HH It is an example of the table which has determined the value of P. It is a flow of MQ encoding processing in consideration of the number of encoding passes to be discarded. FIG. 8 is a partial extraction example of a truncation table stored in a memory in an image encoding device according to a second embodiment of the present invention. Example of selection result prediction truncation table in the image coding apparatus according to the second embodiment of the present invention, in which an offset value (ΔT) to the table used in the selection result prediction next is added to the truncation table of FIG. It is. In the operation flow of prediction result prediction in the image coding apparatus according to the second embodiment of the present invention, in which an offset value (ΔT) up to the table used for prediction result selection is added to the truncation table of FIG. is there. It is a block diagram of the image coding apparatus which concerns on the 2nd Embodiment of this invention provided with another table for selection result prediction. It is an example of the table for prediction in the image coding apparatus which concerns on the 2nd Embodiment of this invention provided with another table for selection result prediction. It is an operation | movement flow of selection result prediction in the image coding apparatus which concerns on the 2nd Embodiment of this invention provided with another table for selection result prediction. It is an example of the table for prediction in the image coding apparatus which concerns on the 2nd Embodiment of this invention provided with another table for selection result prediction. An image code according to the second embodiment of the present invention, in which a prediction value is obtained by approximately calculating the number of paths to be discarded using a predetermined calculation formula instead of using a table in selection result prediction. FIG. FIG. 15 is an operation flow of selection result prediction in the image encoding device shown in FIG. 14. FIG. FIG. 16 (1) is an equation of a straight line graph used in the operation flow of selection result prediction shown in FIG. FIG. 16B is an integerized arithmetic expression used in the operation flow of selection result prediction shown in FIG. It is a block diagram of the image coding apparatus provided with the code amount control means of the prior art. FIG. 18 (1) shows the result of wavelet transform performed independently on luminance (Y) data and color difference (Cb, Cr) data when the image data is that of a color image. It is a figure which shows a mode that it stores in the memory 2 for every subband in a component. FIG. 18 (2) shows that image data is converted into coefficient data for each subband by two-stage wavelet transform (division level 5) by two-dimensional wavelet transform, and each subband is a code block of 32 × 32 size. It is a figure which shows the example divided | segmented into. FIG. 18 (3) is a diagram illustrating the concept of bit plane division. In FIG. 18 (4), bit data of one bit plane is divided into sub-bit planes called three paths depending on the influence on image quality at the time of decoding, and code is given priority over data belonging to a path that seems to have a large influence. It is a figure which shows a mode that conversion is performed. The MQ code that is the result of the MQ encoding is stored in the memory for each code block included in each component and each subband, and the number of zero bit planes of the code block as information necessary for encoding A state in which (#ZBP), the number of encoded paths (#CP), and the code amount (len) of the MQ code for each path are stored in the memory is shown. FIG. 20 (1) is a more detailed block diagram of the encoding control / bitplane dividing unit, MQ encoding unit, and code amount calculation unit. FIG. 20 (2) is a control flow of the encoding control / bit plane dividing unit. FIG. 20 (3) shows the correspondence between actual subband names and numbers. FIG. 3 is a schematic circuit diagram of a coded bit plane number calculation circuit. It is a processing flow of MQ coding in units of code blocks. FIG. 23A is a diagram of an internal circuit of the code amount calculation unit. FIG. 23B is a diagram illustrating a state in which the code reduction amount data is stored in the memory for each component, each subband, and each path. It is an example of a truncation table. It is an example of the control flow of a rate control circuit part.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings.

<< First Embodiment >>
FIG. 1 is a block diagram of an image coding apparatus including a table selection result prediction unit according to the first embodiment of the present invention. In the present embodiment, the selection result prediction circuit 28 uses the result of the code amount calculation unit 16 and the truncation table 8 stored in the memory 2 at the end of the MQ coding of a predetermined subband. In a typical code stream, a path expected to be discarded is predicted, and the encoding control unit 24 performs MQ encoding of the remaining subbands based on the prediction result.

  FIG. 2 shows an operation flow of the encoding control unit 24 at this time. This flow is a wavelet transform of a color image at a division level of 5, and subbands (SB = 1, 2,...) Of 5LL to 3HH for each component of Y, Cb, Cr (C = 1, 2, 3). The selection result prediction is performed when the MQ encoding of 10) is completed, and the remaining 2HL to 1HH subbands are subjected to MQ encoding based on the prediction result.

The first half of the flow indicated by “A” in FIG. 2 is a flow of MQ encoding from 5LL to 3HH subbands with each component of Y, Cb, and Cr. The content of the operation is the same as that of the prior art shown in FIG. The latter half of the flow indicated by “B” in FIG. 2 is a flow for predicting the selection result and MQ coding the remaining subbands based on the result. That is, the operation flow of FIG. 2 and the operation flow of FIG. 20 (2) are [1] selection result prediction at the time when MQ coding from 5LL to 3HH sub-bands is completed for each component Y, Cb, and Cr. Process (S218) is included, and
[2] The remaining 2HL to 1HH subbands are different in that encoding processing is performed in consideration of the number of encoding passes to be discarded, which is a prediction result. Hereinafter, these two points will be described.

First, the selection result prediction process (S218) will be described.
Here, even if the paths described in C = 1, 2, 3, and SB = 1 (5LL) to 10 (3HH) are discarded, the code amount does not become smaller than a target value, and SB = 1 A table having the largest value of (5LL) to 10 (3HH) and the smallest value of SB = 11 (2HL) to 16 (1HH) is selected, and SB = 11 (2HL) to 16 (1HH) of the table is selected. Is the predicted value. FIG. 3 schematically shows the above relationship. Here, it is assumed that MQ coding is finished for subbands from 5LL to 3HH, and the total code amount in the shaded area exceeds a certain target value. The table numbers Ta to Tb have the same contents from 5LL to 3HH, the Tb + 1 table has a smaller code amount than the target value, and the Ta-1 table has different contents from 5LL to 3HH. The Ta table is selected.

ＴＰ、ＴＰＢは、全てのコンポーネント、サブバンドで破棄するパスが０となるように初期化される（Ｓ４０２）。 FIG. 4 shows an operation flow for predicting the selection result. The variables in FIG. 4 have the following contents, respectively.

TP and TPB are initialized so that the path to be discarded in all components and subbands is 0 (S402).

  In the flow of FIG. 4, after starting and initializing each variable (S402), a prediction code amount (Nsub) used for selection result prediction is calculated from the set code amount (N) (S404). As a calculation method, the ratio of the number of wavelet coefficients of the entire image and the number of wavelet coefficients of the subband after MQ coding can be used. In the present embodiment, the selection result prediction is performed at the time when the sub-band encoding up to 3HH is finished at the division level 5, but the number of wavelet coefficients of the sub-band after the encoding is 1/16 of the whole, It is also possible to set the prediction code amount to N × P = 1/16 × N, or to calculate the prediction code amount using the number and average value of coefficient data of each subband. In the table of FIG. 5, the number of coefficient data of each subband and the average value thereof, and further, a weight is set for each subband and multiplied for each subband, and the total of 5LL to 3HH and 2HL to The value of P is determined by the total ratio of 1HH.

  Returning to FIG. 4, first, the total code amount (Q) and the predicted code amount (Nsub) are compared (S406). If the total code amount (Q) is smaller, the selection result prediction process is terminated. At this time, since the value of TPB is set to Tsub (S432), the number of paths to be discarded becomes 0, and all paths are encoded in the subsequent MQ encoding.

In S406, when the total code amount (Q) is large, the following processing is repeated.
[1] The truncation table with the table number T is read from the memory 2 and set to TP (S408).
[2] In the range of C = 1, 2, 3, and SB = 1 (5LL) to 10 (3HH), it is confirmed whether there is a difference between the values of TP and TPB (S410).
[3] When there is a difference between TP and TPB,
[3-1] The number of paths discarded in subbands having a difference is set as a path number CP (S412).
[3-2] A code reduction amount (Σlen (C, SB, CP)) for each path of the corresponding path is subtracted from Q (S414).
[3-3] Q and the predicted code amount (Nsub) are compared (S416). If Q is smaller, the value of TPB is set to Tsub (S432), and the process is terminated.
[3-4] If Q is larger, T is incremented by 1, the value of TP is set to TPB (S420), and the process returns to [1] above.
[4] If there is no difference between TP and TPB, T is incremented by 1 (S426), and the process returns to [1] above.

Next, FIG. 6 is a flow of MQ encoding processing in consideration of the number of encoding passes to be discarded. Each variable and operation contents are as follows.

In the flow of FIG. 6, after the start, the initial values of BN and CP and Cplast are determined using Bp and Tsub (S602). If CPlast is greater than the initial CP value (YES in S604), there is no path to be encoded in this code block, and the flow ends.
When CP is large (S604 / NO), the following procedure is repeated.
[1] The bit plane data of the bit plane number BN is read from the bit plane division memory, and is MQ encoded by the MQ encoder 26 (cleanup pass) (S606).
[2] Pend is asserted when encoding of the cleanup pass is completed (S608).
[3] If the processed path (CP) is the last path (Cplast) (S610, YES), the process ends.
[4] CP and BN are decremented (S612).
[5] The bit plane data of the bit plane number BN is read from the bit plane division memory, and is MQ encoded by the MQ encoder 26 (significance propagation path) (S614).
[6] Pend is asserted when the encoding of the signature propagation path is completed (S618).
[7] If the processed path (CP) is the last path (Cplast) (S620 / YES), the process ends.
[8] The CP is decremented (BN is not changed) (S622).
[9] The bit plane data of the bit plane number BN is read from the bit plane division memory, and is MQ encoded by the MQ encoder 26 (magnitude refinement pass) (S624).
[10] Pend is asserted at the point when encoding of the magnesium refinement path is completed (S626).
[11] If the processed path (CP) is the last path (Cplast) (YES in S628), the process ends.
[12] The CP is decremented (BN is not changed) (S630), and the process returns to [1] (S606).

  With the above processing, in the MQ encoding of the latter half “B” of the flow in FIG. 2, it is possible to prevent encoding in advance for a path that is expected to be discarded in the final code stream.

<< Second Embodiment >>
In the truncation table 8 used for code amount control, the number of paths to be discarded for all subbands is described, but the paths to be discarded in the subband used in the selection result prediction between the two preceding and following tables. There may be no difference in number. This confirms that there is a difference between the values of TP and TPB in the range of C = 1, 2, 3, and SB = 1 (5LL) to 10 (3HH) as in the operation flow of FIG. If there is no difference, it is necessary to read the next table without using that table, and it takes time to predict the selection result.

  This will be described in detail with reference to the table example of FIG. The table example of FIG. 7 is a part of the truncation table example stored in the memory 2. From T196 to T197, the number of paths to be discarded by Cb4LH increases by 1. Thereafter, from T197 to T207, the subbands of C = 1, 2, 3, and SB = 1 (5LL) to 10 (3HH) have the same value. . Compared to T207, the number of paths discarded at T208 increases by 1 at Y1LH, and C = 1, 2, 3, and SB = 1 (5LL) to 10 (3HH) have the same value. When the selection result prediction is performed using this table, from T198 to T207, it is necessary to perform an operation of reading from the memory even though it is not actually used, and an operation of comparing the set number of paths with the previous table. .

  The selection result prediction operation flow shown in FIG. 9 and the table shown in FIG. 8 add an offset value (ΔT) to the table used in the selection result prediction to the truncation table 8 in FIG. Thus, in the method for solving the above problem, there are an operation flow and a truncation table example of selection result prediction according to the second embodiment of the present invention.

  In the table of FIG. 8, T197 includes an offset value (ΔT) up to T209 that has a difference in the number of paths in subbands C = 1, 2, 3, and SB = 1 (5LL) to 10 (3HH). = 12 is given. In the flow of FIG. 9, the table number is not simply incremented (see FIG. 4, S420), but an offset value (ΔT) is added (S920). As a result, the next table to be read has a difference of 1 in the number of sub-band paths to be subjected to selection result prediction compared to the current table, so unnecessary reading of the table does not occur.

  FIG. 10 is a block diagram of an image coding apparatus that solves the above problem by having a separate table for prediction of selection results. In this example, the selection result prediction circuit 28 uses a prediction table stored in the memory 3 (prediction table) 30 instead of the truncation table 8 of the memory 2. FIG. 11 shows an example of the prediction table, and FIG. 12 shows an operation flow of selection result prediction in this example. As shown in FIG. 11, the prediction table describes the paths to be discarded in each subband of C = 1, 2, 3, and SB = 5LL to 3HH. The value of the number of paths to be discarded is always different by 1 in each subband. In addition, the truncation table stored in the memory 2 and the number (TA) of the table with the same number of discarded paths with C = 1, 2, 3, SB = 5LL to 3HH and the smallest number are also described. ing.

The following variables are added to the selection result prediction operation flow shown in FIG.

  The truncation table 8 stored in the memory 2 is read using the TA value described in the finally selected table (TPB ← TBL (TAB)) (S1226), and the subband to be encoded thereafter. (Tsub (C, SB) ← TPB (C, SB)) (S1228). Further, a prediction table with no TA value description as shown in FIG. 13 can also be used. At this time, the number of 3HH discarded paths of each component described in the finally selected table may be the number of paths discarded in the subband to be encoded later.

Furthermore, instead of using a table for prediction of the selection result, it is also possible to obtain a predicted value by approximately calculating the number of paths to be discarded using a calculation formula given in advance. FIG. 14 is a block diagram of an image coding apparatus that performs selection result prediction according to this method. FIG. 15 shows an operation flow of selection result prediction in the image encoding device shown in FIG. In this example, yt = K × (x−10) + t (t = 1, 2, 3,...) Shown in FIG.
The selection result prediction is performed using the linear formula and the integer calculation formula of FIG. In FIG. 16A, “K” indicates the slope of a straight line, and a predetermined value is given. Further, “floor [yi]” in FIG. 16B represents the maximum integer that does not exceed “yi”. However, in order to simplify the description here, the component is only Y. Yt (SB) is the number of paths to be discarded. By incrementing t = 1, 2,..., The number of paths to be discarded increases even if the SB is the same.

In the operation flow shown in FIG. 15, the basic flow after initialization is as follows.
[1] Yt (SB) is used as a calculation formula.
[2] It is checked whether there is a difference between the values of TP (SB) and Yt (SB) in any subband (S1510).
[3] When there is a difference between TP and Yt (S1510 / YES),
[3-1] Let Yt (SB) of subbands with differences be TP (SP) and CP (S1512).
[3-2] The code reduction amount (Σlen (SB, CP)) for each path of the corresponding path is subtracted from Q (S1514).
[3-3] Q and the predicted code amount (Nsub) are compared (S1516). If Q is smaller (S1516, NO), the value of TPB is set to Tsub, and the process ends.
[3-4] When Q is larger (YES in S1516), the process proceeds to [4] below.
[4] When there is no difference between TP and TPB (S1510, NO), or when Q is larger in (3-4) (S1516, YES),
[4-1] If SB ≠ 1, SB is decremented (S1520), and the process returns to [1] above.
[4-2] If SB = 1 and t is already at the maximum value (Tmax) (S1522 / YES), the process is terminated. Otherwise (S1522 / NO), t is incremented (S1524). ) Return to [1] above.

  Thereafter, the predicted value (path discarded in the subband to be encoded) is Yt−1 (SB), and SB = 11 (2HL), 12 (2LH),... 16 (1HH) is calculated. Is required.

DESCRIPTION OF SYMBOLS 2 ... Memory, 12 ... Wavelet conversion part, 16 ... Code amount calculation part, 18 ... Rate control circuit part, 20 ... Packet generation circuit, 22 ... Code formation part, 24 ... Encoding control unit, 26... MQ encoding unit, 28.

Japanese Patent No. 4064279 Japanese Patent No. 40003628 JP 2005-167611 A

Claims (5)

A frequency conversion means for converting the frequency of the image data;
Bit plane encoding means for encoding the coefficient data after frequency conversion into a rectangular area called a code block for each bit plane after being divided;
Storage means for storing a plurality of tables describing combinations of the number of bit planes to be discarded with respect to the code after bit plane encoding;
In an image encoding device including code amount control means for controlling a code amount by selecting a set of tables that are equal to or less than a predetermined code amount from the plurality of tables,
Furthermore, table selection result prediction means for predicting a set of tables selected when the entire image is encoded using the code amount at the time when the encoding of the code block included in the predetermined subband is completed; ,
An image coding apparatus comprising coding bit plane control means for controlling coding means so as to determine bit planes to be coded with code blocks included in the remaining subbands based on the prediction result. The image encoding device according to claim 1,
From the set code amount used in the code amount control and the size of the predetermined subband, a prediction code amount assigned to the corresponding subband is calculated,
From the plurality of tables describing combinations of the number of bit planes to be discarded to be used in the code amount control, using the code amount for each bit plane when the encoding of the code block included in the subband is completed, Select the table with the maximum number of discarded bitplanes in the range where the total subband code amount does not fall below the predicted code amount,
An image encoding device that encodes the code blocks included in the remaining subbands using the number of bit planes discarded in the corresponding subband described in the selected table. The image encoding device according to claim 2,
An image coding apparatus for calculating a prediction code amount allocated to the coded subband using an average value of wavelet coefficient data for each subband. The image encoding device according to claim 2,
A plurality of second tables describing a combination of the number of bit planes to be discarded only for the previously encoded subbands and data specifying the table used in the code amount control corresponding to each table. An image encoding device that performs table selection result prediction using the. The image encoding device according to claim 2,
An image encoding apparatus that predicts a table selection result by calculating the number of bit planes to be discarded using a predetermined calculation formula.

Images