JP2005151167A - Encoding apparatus and encoding program for predicting movement compensation - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To suppress the amount of encoding relative to reference indexes to be assigned to reference images in a movement compensation predicting apparatus using a plurality of reference images. <P>SOLUTION: A movement-related information processing part 118 counts the number of each of reference indexes relative to the image to be encoded. The amount of encoding A is obtained under the assignment condition of predetermined encoding bits concerning each reference index, and the amount of encoding B is obtained under the assignment condition where the number of the encoding bits of each reference index is made to correspond to the number of the reference indexes by ascending/descending order relation. In the case of A-B>α, the reference indexes are each changed to meet the assignment condition used in obtaining B and encoded under the condition by an entropy encoder 104. Then a multiplexer 105 incorporates the change information to the high-order hierarchy of multiplex image data. A(α) is defined as the amount of encoding of the change information. In the case of A-B≤α, encoding is performed under the predetermined assignment condition without performing changing. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention is a technique related to motion compensated predictive coding that performs high-efficiency coding of moving picture information, and in particular, when a plurality of reference pictures are used for motion compensated prediction, a reference index indicating each reference picture is used. The present invention relates to an improvement for suppressing the code amount.

  In MPEG (Moving Picture Experts Group), H.263, etc., which are international standards for moving picture coding technology, a hybrid code combining motion compensation prediction and DCT (Discrete Cosine Transform) based on inter-frame prediction coding. The mainstreaming method is widely used in the field of multimedia information communication.

In inter-frame predictive encoding, the image signal of the previous frame stored in the memory is used as a predictive signal for the image signal of the current frame, and a prediction error signal corresponding to the difference between both frames is encoded.
On the other hand, in motion compensation prediction, a motion vector for each block of an image is obtained from the previous frame and the current frame, the previous frame is corrected using the motion vector, and a predicted image for the current frame is generated. A prediction error is obtained by taking a difference from the image of the current frame, and the motion vector and the prediction error signal are encoded.
That is, according to motion compensation prediction, it is possible to reduce the amount of information generated by improving the prediction accuracy of a moving region of an image and improve image quality, which is an essential technique in a high-efficiency encoding method.

  In motion-compensated prediction, for example, there is a method of selecting and encoding a reference image to be used from among a plurality of reference images as shown in Patent Document 1 below. It is necessary to encode a motion vector and a reference index indicating a reference image referred to by the motion vector.

Hereinafter, an encoding apparatus that performs motion compensation prediction using a plurality of reference images as described above will be described.
The coding apparatus has the configuration shown in the functional block diagram of FIG. 3, and a frame image input in time series is subjected to forward predictive coding from an intra-frame coded image (I0 picture) and I0 picture. Select a picture (P1 picture), an I0 picture or a P1 picture, and a picture (P2 picture) that has been forward-predicted and coded, and select any one of the pictures (I0, P1, P2). Are sequentially encoded as a converted image (P3 picture), and output to the transmission path as a bit stream multiplexed with accompanying information and the like obtained in the encoding process.
Note that the reference relationship of each image type in forward prediction encoding is as shown in FIG.

First, in the intra-frame coding mode in which the input frame image 100 is coded into an I0 picture, the switch 101a is connected to the x side and the switch 112a is in a disconnected state.
In this mode, the orthogonal transform process is performed on the input frame image 100 by the orthogonal transform unit 102 in units of 8 × 8 pixel blocks, and the coefficient obtained by the process is given to the quantizer 103. Then, the coefficient data is quantized with a predetermined number of quantization steps, and the fixed length code is output to the entropy encoder 104 and the inverse quantizer 110.

The entropy encoder 104 converts the two-dimensional 8 × 8 coefficients into one dimension and encodes the coefficients by a predetermined method.
The multiplexer 105 multiplexes the incidental information obtained from the quantizer 103 together with the entropy-encoded code string as header information, and outputs the multiplexed information as a coded bit stream 120 to the transmission path.
Since the I0 picture is obtained by intraframe coding processing, the motion detector 117 and the motion compensation predictor 116 do not operate, and there is no additional information such as a motion vector.

  On the other hand, the quantized data output to the inverse quantizer 110 is subjected to the inverse processing of the orthogonal transformer 102 and the quantizer 103 by the inverse quantizer 110 and the inverse orthogonal transformer 111 to obtain the original data. After the input frame image 100 is reproduced and the reproduced image is saved in the image memory 113, the block noise is reduced by the loop filter 114 and saved in the image memory 115.

Next, in a mode in which the input frame image 100 is encoded into a forward prediction encoded image (P1 picture, P2 picture, P3 picture), the switch 101a is connected to the y side, and the switch 112a is in a connected state.
In this encoding mode, the input frame image (current frame image) 100 is input to the motion detector 117 and the subtractor 101.
The motion detector 117 searches for a correlation with the reference image saved in the image memory 115 for each target block (16 × 16 pixels or 8 × 8 pixels) obtained by dividing the input frame image 100 to perform motion. A vector is obtained, and a reference index related to the reference image for which the motion vector is obtained is assigned.
However, detailed functions of the motion detector 117 and the motion compensation predictor 116 will be described later. Here, processing subsequent to the subtracter 101 will be described first.

In the subtractor 101, a difference between the image signal of the target block of the input frame image 100 and the inter-picture prediction signal obtained from the motion compensation predictor 116 is obtained, and an orthogonal transformer is obtained as a prediction error signal via the switch 101a. 102.
The processing performed by the orthogonal transformer 102 and the quantizer 103 is only to obtain the prediction error signal and the image data that has been subjected to the forward prediction encoding, and basically the intra-frame processing is performed. This is the same as the encoding mode.
However, when the encoded bit stream 120 is formed by encoding and multiplexing the encoded image data and the auxiliary information by the entropy encoder 104 and the multiplexer 105, along with the auxiliary information obtained from the quantizer 103, The additional information such as the motion vector obtained by the motion detector 117 and the reference index is also encoded and multiplexed as header information.

On the other hand, the quantized coefficient data input from the quantizer 103 to the inverse quantizer 110 side is processed by the orthogonal transformer 102 and the quantizer 103 by the inverse quantizer 110 and the inverse orthogonal transformer 111. Since the inverse process is performed, the prediction error signal that is the output of the subtractor 101 is restored.
Since the switch 112a is in the connected state, the prediction signal of the motion compensation predictor 116 is added to the restored prediction error signal by the adder 112 to reproduce the image of the original target block, and the reproduced image is the image. It will be saved in the memory 113.

  The procedure is sequentially executed for each target block in one frame, and when reproduced images are obtained for all target blocks for one frame, the image data in the image memory 113 is read out, and block noise is detected by the loop filter 114. The image is reduced and saved in the image memory 115.

  Incidentally, the image memory 115 is a memory that stores a maximum of three frames of reproduced images in a first-in first-out (FIFO) system, and includes one intra-frame coding mode and three forward prediction coding modes. Are alternately set, the P1 picture, the P2 picture, and the P3 picture corresponding to the previous three frames are encoded in the I0 picture, and the previous three frames are encoded in the P1 picture. The P2 picture, the P3 picture, and the I0 picture are related to the P2 picture, the P3 picture is encoded with the P3 picture, the I0 picture, and the P1 picture that are immediately before the P3 picture. The reproduced images of the I0 picture, the P1 picture, and the P2 picture related to the three frames are saved.

Here, the functions of the motion detector 117 and the motion compensation predictor 116 that have been briefly described above will be described in detail.
The motion detector 117 and the motion compensated predictor 116 operate only in the forward predictive encoding mode. As shown in FIG. 4, only the reproduced image of the I0 picture is encoded at the time of encoding into the P1 picture. When encoding to a P2 picture, a playback image of the I0 picture or P1 picture is used as a reference image, and when encoding to a P3 picture, any of the playback images of the I0 picture, P1 picture, or P2 picture is used as a reference image.
Then, the motion detector 117 selects a reference image under the above conditions from among three frames of reproduced images (candidate reference images) saved in the image memory 115 according to the encoded image type in the forward predictive encoding mode. To do.

For example, when encoding into a P3 picture, three candidate reference images (reproduced images of an I0 picture, a P1 picture, or a P2 picture) are the same size as the target block on the input frame image 100 side (16 × 16 pixels or 8 × 8 pixels) ), And for each target block of the input frame image 100, each reference block in each candidate reference image is ½ pixel within a certain range (for example, ± 16 pixels in the case of 16 × 16 pixels). 15 pixels) is moved to perform block matching with the target block, and the movement of the reference block having the smallest absolute error sum (or square error sum) is obtained as a motion vector. Assign the associated reference index.
That is, the motion detector 117 obtains a motion vector based on the above condition for each target block of the input frame image 100, and gives a reference index of the reference image used for obtaining the motion vector.
In that case, the reference index is associated with the P3 picture in order from the closest to the time, and as shown in FIG. 4, “0” is assigned to the reference images related to the pictures P2, P1, and I0. , “1” and “2” are associated with each other.
Further, as shown in the following Table 1, code bits “0”, “10”, and “110” are assigned to each reference index, respectively.

In the above description, the encoding to the P3 picture has been described. However, the method for obtaining the motion vector is the same when encoding to the P1 picture or the P2 picture.
However, since the reference picture at the time of encoding to the P1 picture is only the reproduced picture of the I0 picture, the reference index is only “0” for the reference picture, and the reference picture at the time of encoding to the P2 picture is the I0 picture. And P1 picture reproduction images, the reference indexes “1” and “0” are associated with each other.

Next, the motion compensation predictor 116 moves the pixel block of the size of each candidate reference image saved in the image memory 115 on the basis of the motion vector and the reference index given from the motion detector 117. Inter prediction signal is obtained.
Then, as described above, the subtractor 101 calculates the difference (prediction error) between the inter-picture prediction signal and the input image signal and outputs it to the orthogonal transformer 102.
Also, the motion vector obtained by the motion detector 117 and the additional information of the reference index as described above are transferred to the entropy encoder 104 and the multiplexer 105 and encoded based on a predetermined rule (the reference index is shown in Table 1). (Encoding by rule) and multiplexing are performed and output as an encoded bit stream 120 in the same manner as described above. However, when assigning code bits to a reference index in the entropy encoder 104, the above table is used. As shown in FIG. 1, “0”, “10”, and “110” are assigned to the reference indexes “0”, “1”, and “2”, respectively.

JP 11-239351 A

By the way, in the conventional motion compensated predictive coding apparatus, the motion detector 117 associates the reference index value with the reference image that is temporally closer to the input image in ascending order, and the entropy encoder. In 104, the smaller the reference index value is, the smaller the number of coded bits is assigned.
This is because, in moving images, images that are closer in time generally have higher correlation, and in motion compensated prediction, a reference image that is closest in time to an encoding target image is often selected. This is because the encoding amount of the reference index can be reduced by encoding based on the allocation condition as described above.

However, for example, in a moving image in which the entire screen is periodically flashed or a moving image in which there is a fine movement due to camera shake or the like, it is not a reference image immediately before the encoding target image but another reference image. May have high correlation.
In such a case, if the correspondence of the reference index to the reference image is uniformly applied, there is a disadvantage that the code amount of the reference index becomes unexpectedly large.

  Therefore, the present invention provides a motion compensated predictive coding apparatus that reduces the code amount related to the reference index by adaptively changing the association relationship of the reference index with the reference image to change the coding bit allocation condition. It was created for the purpose of providing.

  The present invention divides an encoding target image into target blocks of a predetermined size, reversely processes image data encoded before the encoding target image, and reproduces a plurality of reference images. In a mode in which an encoded image is generated with reference to two or more reference images, a motion vector is detected by obtaining a closest approximation from the blocks in the reference image for each target block, and a motion vector for each target block is detected. And inter-image prediction data by motion compensated prediction using the reference image, and then obtaining a prediction error that is a difference between the inter-image prediction data and the encoding target image, and for each reference image Each reference index that is an ascending order from the temporally closer to the encoding target image is assigned, and a code for each reference index is assigned to the ascending order And a motion compensated predictive coding apparatus for outputting multiplexed image data having a hierarchical structure in which the prediction error, each motion vector, and each reference index are encoded, so as to be allocated under a condition that the code amount increases accordingly. A counting means for counting the number of each reference index relating to the encoding target image; and a multiplication value of the number of each reference index obtained by the counting means and the code amount of the code assigned to each reference index. Multiplication when first calculation means for calculating the sum, the number of each reference index obtained by the counting means and the code amount of the code assigned to each reference index are associated in an ascending / descending order inverse relationship A second calculation means for calculating a sum of values, and a difference obtained by subtracting the value obtained by the second calculation means from the value obtained by the first calculation means. If greater than the value, the number of each reference index is changed to an ascending order number given in descending order of the number of each reference index obtained by the counting means, and if the difference is less than or equal to the predetermined value, Reference index control means for maintaining each reference index as a predetermined assigned number, and when the reference index control means changes the number of each reference index, the change information is stored in the upper hierarchy of the multiplexed image data. The present invention relates to a motion-compensated predictive coding apparatus characterized by comprising setting information coding means for performing coding.

In the motion compensated predictive coding apparatus of the present invention, the number of each reference index related to the encoding target image is obtained by the counting means, and the first calculation means assigns the reference index of each reference picture and each reference index according to a predetermined condition. Based on the relationship with the obtained code amount, the code amount related to all the reference indexes is estimated, and the number of each reference index obtained by the second calculating means and the code amount of each reference index are calculated in ascending / descending order. The amount of codes related to all reference indexes in the case of association with the inverse relationship is calculated.
The reference index control means determines the reference index number when the code amount obtained by the second computing means is smaller than the code quantity obtained by the first computing means and the difference is larger than a predetermined value. By changing to the ascending order number given in descending order of the number of each reference index obtained by the counting means, the code amount related to the reference index becomes the code quantity obtained by the second computing means, and vice versa The code amount obtained by the first calculation means is set without changing the number of each reference index.
In other words, when detecting a motion vector related to a target block, a code having a smaller code amount is assigned to a reference index of a reference image that has been referred to more frequently, thereby reducing the total coding amount for all reference indexes. ing.
Further, such setting of assignment can be determined by encoding the change information in the upper layer of the multiplexed image data by the setting information encoding means.
The “predetermined value” is reasonable if it is set as a value that is greater than or equal to the amount of code when the setting information encoding means encodes the change information, and when the value fluctuates under various conditions. May adopt the maximum value or the average value of the code amount.

  The processing of each means in the invention can be executed by a computer, and the program in that case includes a first procedure for counting the number of each reference index related to the encoding target image, and each of the steps obtained in the first procedure. A second procedure for calculating a sum of multiplication values of the number of reference indexes and the code amount of the code assigned to each reference index; the number of each reference index obtained in the first procedure and the assignment to each reference index A third procedure for calculating the sum of multiplication values in which the code amount of the code is associated in an ascending / descending order inverse relationship, and the value obtained in the third procedure from the value obtained in the second procedure. If the subtracted difference is larger than a predetermined value, the number of each reference index is changed to an ascending order number given in descending order of the number of each reference index obtained in the first procedure, If the difference is less than or equal to the predetermined value, the reference number remains the default assigned number, and the change is made when the number of each reference index is changed in the fourth step. A fifth procedure for encoding information into the upper layer of the multiplexed image data is executed.

  The present invention provides a motion-compensated predictive coding apparatus that uses a plurality of candidate reference images, and assigns a small number of coding bits to a reference index having a high reference frequency and a code amount related to a reference index indicating each reference image. Trial calculation is performed according to the conditions, and by changing the reference index adaptively, the code bit allocation condition for each reference index is changed to reduce the code amount related to the reference index.

Hereinafter, an embodiment of a motion compensated prediction encoding apparatus according to the present invention will be described in detail with reference to FIGS. 1 and 2.
FIG. 1 is a functional block circuit diagram of the encoding apparatus according to this embodiment. As is apparent from a comparison between FIG. 3 and FIG. 3, in the circuit for transferring incidental information from the motion detector 117 to the multiplexer 105. The only difference is that the motion-accompanying information processing unit 118 is added.
As described above, when the input frame image (encoding target image) is encoded into the P2 picture or the P3 picture, the motion detector 117 performs the processing for each target block (16 × 16 pixels or 8 × 8 pixels). When a motion vector is obtained, which reference image is selected is indicated by a reference index. The feature of this encoding device is that each of the motion-accompanying information processing unit 118 and the entropy encoder 104 Checking the selection state of the reference image, changing the reference index, and changing the allocation condition of the encoded bit for each reference index. Other data processing operations are the same as the encoding apparatus described in the background section. It is the same.
Therefore, the operation related to the characteristic function will be mainly described here, and description of other data processing operations will be omitted.

The flowchart of FIG. 2 shows an operation procedure in which the motion-accompanying information processing unit 118 and the entropy encoder 104 execute the function.
However, this flowchart relates to a case where an input frame image is encoded into a P3 picture. As described in the background art section, the image memory 115 has three reproduced images of the I0 picture, the P1 picture, and the P2 picture. Saved as a candidate reference image.

First, the motion detector 117 obtains a motion vector and a reference index based on the reference image related to the reference image for which the motion vector is obtained for each target block of the input frame image. Those data relating to all target blocks are temporarily saved in the internal memory (S11).
Then, the motion-accompanying information processing unit 118 counts the number N (i) of each reference index [i] (i = 0, 1, 2) (S12).

  The condition for assigning the number of encoded bits B (i) to each reference index [i] when the input frame image is encoded into a P3 picture is given by the correspondence shown in Table 1 above. The unit 118 uses the number N (i) of each reference index [i] and the number of encoded bits B (i) to code the amount of code A = N (when all reference indexes are encoded under a predetermined condition. 0) * B (0) + N (1) * B (1) + N (2) * B (2) is estimated (S13).

The motion-accompanying information processing unit 118 obtains the descending order of the number N (i) of each reference index [i], and the order is N (a) ≧ N (b) ≧ N (c) [where a, b, If cε0, 1, 2], the code amount B = N (a) * B (0) + N (b) * B (1) + N (c) * B (2) is obtained (S14, S15). ).
That is, the sum of the multiplication values when the number of encoded bits B (i) of each reference index and the number N (i) of each reference index are associated with each other in ascending order and descending order is calculated. Based on the code amount B, a trial calculation is performed.

Next, the motion-accompanying information processing unit 118 obtains a value obtained by subtracting the code amount B from the code amount A, and determines whether the value is larger than a predetermined value α (S16).
Here, α corresponds to a code amount in the case of encoding reference index change information in the entropy encoder 104 described later.

If a comparison result of (A−B)> α is obtained, the reference index [i] is changed, and after changing 0 to a, 1 to b, and 2 to c, each target The motion vector related to the block, each changed reference index [i] (i = a, b, c), and reference index change notification information are output to the entropy encoder 104 (S16 → S17, 18).
That is, the motion-accompanying information processing unit 118 changes the reference index [i] (i = 0, 1, 2) so as to correspond to the descending order of the number N (i) and outputs it to the entropy encoder 104. .

Incidentally, since the entropy encoder 104 assigns code bits to the reference index [i] according to the relationship shown in Table 1, the reference index [i] (i = a, b, c) after the change is used. When the allocation is performed, the encoding bits “0”, “10”, “110” are allocated in descending order of the count numbers of the three types of reference indexes, and the reference indexes related to all the target blocks are encoded. .
In other words, a small number of encoded bits is assigned to a reference image with a high reference frequency (reference index [i] with a large number of counts), and the code amount of the entire reference index can be reduced accordingly.

Specifically, for example, if the number of target blocks referring to the reference images related to the I0 picture, the P1 picture, and the P2 picture are 120, 30, and 70, respectively, the code amount A and the code amount B are respectively Given by
A = 70 * 3 + 30 * 2 + 120 * 1 = 390 (bit)
B = 120 * 1 + 70 * 2 + 30 * 3 = 350 (bit)
Changing the reference index is more advantageous in reducing the amount of codes.
However, when the reference index change notification information is encoded by the entropy encoder 104, the code amount α is added to the code amount of the entire input frame image, so that (AB)> The reference index is changed under the condition of α.
In this case, if α is 10 bits, the reference index is changed, and the code amount for A− (B + α) = 30 bits can be reduced.

  Then, the entropy encoder 104 allocates encoded bits to the changed reference index as described above, encodes it together with motion vectors and other incidental information, and the multiplexer 105 generates the hierarchical bit stream 120. According to the change notification information, the multiplexer 105 encodes the assignment condition change information in the upper layer (for example, the sequence layer, the picture layer, and the slice layer) of the multiplexed image data based on the change notification information. (S19).

  On the other hand, if (A−B) ≦ α in step S16, the motion-accompanying information processing unit 118 outputs the saved motion vector and the reference index [i] to the entropy encoder 104 as they are, and performs entropy encoding. The unit 104 performs encoding under a predetermined allocation condition based on the correspondence relationship between the reference index [i] (i = 0, 1, 2) in Table 1 and encoded bits (S16 → S20, S21).

  As described above, the case where the input frame image is encoded into the P3 picture has been described. However, even when the input frame image is encoded into the P2 picture, the reproduced image of the I0 picture or the P1 picture is used as a candidate reference image. The code amount can be reduced by performing the reference index changing process according to the above procedure in the motion-related information processing unit 118 and the entropy encoder 104.

その場合、ビン列の長さは必ずしもビット数とは一致しないが、動き付帯情報処理部１１８は、上記の場合と同様に、符号量Ａ＝Ｎ(0)＊Bin(0)＋Ｎ(1)＊Bin(1)＋Ｎ(2)＊Bin(2)として試算し、各参照インデックス［ｉ］に係る個数Ｎ(i)の降順を求め、その順序がＮ(a)≧Ｎ(b)≧Ｎ(c)［但し、ａ,ｂ,ｃ∈０,１,２］であれば、符号量Ｂ＝Ｎ(a)＊Bin(0)＋Ｎ(b)＊Bin(1)＋Ｎ(c)＊Bin(2)を求め、以下同様の処理を行う。
更に、より正確な符号量を試算する場合には、前記のような演算ではなく、実際に算術符号化を行った結果に基づいて参照インデックスの並び替えを行うか否かを決定するような方法も採用できる。 When the entropy encoder 104 is an arithmetic encoder in the motion compensated predictive encoding apparatus, a bin sequence as shown in Table 2 below can be assigned to each reference index.

In this case, the length of the bin string does not necessarily match the number of bits, but the motion-accompanying information processing unit 118 performs the code amount A = N (0) * Bin (0) + N (1) as in the above case. * Bin (1) + N (2) * Bin (2) is calculated and the descending order of the number N (i) related to each reference index [i] is obtained, and the order is N (a) ≧ N (b) ≧ N (c) [where a, b, c ∈ 0, 1, 2], the code amount B = N (a) * Bin (0) + N (b) * Bin (1) + N (c) * Bin (2) is obtained, and the same processing is performed thereafter.
Furthermore, when calculating a more accurate code amount, a method for determining whether or not to reorder the reference indexes based on the result of actual arithmetic coding, instead of the above-described calculation. Can also be adopted.

In the column of this embodiment and the background art, the motion compensated predictive coding apparatus is represented by functional blocks as shown in FIGS. 1 and 3, but the whole is represented by a micro signal processor such as a DSP (Digital Signal Processor). It is also possible to configure with a computer circuit and to execute the procedure of FIG. 2 by a program.
The program in that case may be provided not only by a recording medium but also by a method of downloading via a communication network such as the Internet.

It is a functional block circuit diagram of a motion compensation predictive coding apparatus according to an embodiment of the present invention. It is a flowchart which shows the operation | movement procedure of a motion attendant information processing part and a multiplexer. It is a functional block circuit diagram of the motion compensation predictive coding apparatus according to the prior art. It is a figure which shows the reference relationship of each image type in motion compensation prediction encoding.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 ... Input image (input frame image), 101 ... Subtractor, 101a, 112a ... Switch, 102 ... Orthogonal transformer, 103 ... Quantizer, 104 ... Entropy encoder, 105 ... Multiplexer, 110 ... Inverse quantization 111: inverse orthogonal transformer, 112 ... adder, 113, 115 ... image memory, 114 ... loop filter, 116 ... motion compensation predictor, 117 ... motion detector, 118 ... motion-accompanying information processing unit, 120 ... sign Output (encoded bitstream).

Claims (2)

The encoding target image is divided into target blocks of a predetermined size, the image data encoded before the encoding target image is inversely processed to reproduce a plurality of reference images, and two or more references from among them are reproduced. In a mode in which an encoded image is generated with reference to an image, a motion vector is detected by obtaining a closest approximation from the blocks in the reference image for each target block, and the motion vector and the reference image for each target block are detected. After obtaining inter-picture prediction data by motion compensated prediction using the above, a prediction error that is a difference between the inter-picture prediction data and the encoding target image is obtained, and the encoding target picture is obtained for each reference picture In addition to assigning each reference index that becomes an ascending order from the one closer to the time, a code for each reference index is associated with the ascending order number. As allocated under the condition that the amount of issue is large, in the prediction error and the motion compensated predictive coding apparatus which outputs the multiplexed image data of the hierarchical structure and the respective reference indices and each motion vector is coded,
Counting means for counting the number of each reference index relating to the encoding target image;
First computing means for computing a sum of multiplication values of the number of each reference index obtained by the counting means and the code amount of the code assigned to each reference index;
Second computing means for computing the sum of multiplication values when the number of each reference index obtained by the counting means and the code amount of the code assigned to each reference index are associated with each other in ascending / descending order. When,
When the difference obtained by subtracting the value obtained by the second computing means from the value obtained by the first computing means is larger than a predetermined value, the number of each reference index obtained by the counting means is determined as the number of each reference index. Change to the ascending order number given in descending order of the reference index control means, if the difference is less than or equal to the predetermined value, the reference index control means to leave each reference index as a default assignment number;
And a setting information encoding unit that encodes the change information in an upper layer of the multiplexed image data when the reference index control unit changes the number of each reference index. Compensated predictive coding apparatus. The encoding target image is divided into target blocks of a predetermined size, the image data encoded before the encoding target image is inversely processed to reproduce a plurality of reference images, and two or more references from among them are reproduced. In a mode in which an encoded image is generated with reference to an image, a motion vector is detected by obtaining a closest approximation from the blocks in the reference image for each target block, and the motion vector and the reference image for each target block are detected. After obtaining inter-picture prediction data by motion compensated prediction using the above, a prediction error that is a difference between the inter-picture prediction data and the encoding target image is obtained, and the encoding target picture is obtained for each reference picture In addition to assigning each reference index that becomes an ascending order from the one closer to the time, a code for each reference index is associated with the ascending order number. The computer functions as a motion-compensated predictive coding apparatus that outputs multiplexed image data having a hierarchical structure in which the prediction error, each motion vector, and each reference index are encoded, so that the prediction error, the motion vector, and the reference index are encoded. A program for
A first procedure for counting the number of each reference index related to the encoding target image;
A second procedure for calculating a sum of multiplication values of the number of each reference index obtained in the first procedure and the code amount of the code assigned to each reference index;
A third procedure for calculating a sum of multiplication values in which the number of each reference index obtained in the first procedure and the code amount of the code assigned to each reference index are associated in an ascending / descending order inverse relationship;
When the difference obtained by subtracting the value obtained in the third procedure from the value obtained in the second procedure is larger than a predetermined value, the number of each reference index is the number of each reference index obtained in the first procedure. Changing to ascending order numbers assigned in descending order, and if the difference is less than or equal to the predetermined value, the fourth procedure leaving each reference index as a default assignment number;
When the number of each reference index is changed in the fourth procedure, the computer executes a fifth procedure for encoding the change information in an upper layer of the multiplexed image data. Predictive coding program.

Images