JP2007228371A - Image encoding apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image encoding apparatus which achieves an excellent encoding in aspects of an compression efficiency and an image quality in a bidirectional prediction or a double prediction. <P>SOLUTION: An image signal and a prediction signal of an encoded subject are input in a subtractor 11, and a difference of both of the signals is calculated. The difference is processed by an orthogonal converter 12, a quantizer 13, and a variable length encoder 14; and is transmitted as an encoding output. A plurality of reference frames obtained by being locally decoded are stored in a frame memory 18. A repetitive motion searching apparatus 19 uses the image signal of the encoded subject and the plural reference frames, and decides a combination of optimum motion vectors for the plural reference frames by means of a repetitive processing. A motion compensator 20 outputs the prediction signal using the decided motion vector and the plural reference frames. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an image encoding device, and more particularly to an image encoding device capable of obtaining high compression efficiency by mixing a plurality of encoding parameters.

  When transmitting and storing image information, compression by encoding is performed to reduce the amount of information. One type of image information encoding is predictive encoding, which reduces redundant image information by motion compensation using inter-frame correlation.

  In predictive coding, high compression efficiency can be obtained by mixing a plurality of coding parameters. For example, in bi-prediction in an MPEG macroblock and bi-prediction in an H.264 macroblock, there are encoding parameters for each reference frame to which an encoding target frame refers.

  In bidirectional prediction, a frame that is temporally before and after the encoding target frame is used as a reference frame, and a motion vector is calculated as an encoding parameter for each macroblock between the encoding target frame and each reference frame. The predicted image referred to by the encoding target frame is obtained by averaging the values of the corresponding pixels in the macroblock at the position indicated by the motion vector.

  In bi-prediction, multiple frames in the same direction in time can be used as reference frames by removing the temporal restriction in bidirectional prediction.

  In Patent Document 1, the forward motion vector and the backward motion vector derived in the forward motion compensation inter-frame prediction and the backward motion compensation inter-frame prediction are respectively distributed in a size corresponding to the ratio of the frame intervals. In addition, a moving picture coding method for deriving a forward motion vector and a backward motion vector for a frame (intermediate frame) that is not a reference frame in inter-frame prediction is described.

  In Patent Document 2, first, a backward motion vector to the backward reference image of the current image block is obtained, and then the backward motion vector and the forward reference image block indicated by the backward motion vector are forwarded. By calculating a new vector from the two motion vectors for the direction reference image and using the calculated vector as the forward motion vector, the forward motion vector and the backward motion vector are reduced with a small amount of calculation. A motion vector search method capable of obtaining the above is described.

  In Patent Document 3, a delta motion vector for optimally predicting a B block is searched for when encoding is predicted from both directions using temporally preceding and succeeding frames, and the delta motion vector is converted into a forward motion vector. Video image coding that can effectively compress even non-linear motion scenes by adding motion vectors to obtain final forward motion vectors and subtracting delta motion vectors from backward motion vectors to obtain final forward motion vectors A method is described.

In Patent Document 4, when bi-directional predictive coding is performed, first, an optimal forward motion vector is obtained, and for the reverse motion vector, an average value of prediction data in both directions is selected from the backward motion vector candidates. And a moving image processing method in which the one showing the highest correlation is selected (third embodiment).
JP 2002-335529 A JP 2000-295625 A Japanese Patent Laid-Open No. 9-182083 JP 2000-308062 A

  In image compression schemes such as MPEG and H.264, in order to increase the compression efficiency of moving objects and scenes by motion compensation, minimizing prediction errors is used as an evaluation criterion when searching for motion vectors. In general, a motion vector search selects a motion vector that minimizes a square error with the position of a reference frame corresponding to the motion vector. The same applies to bidirectional prediction and bi-prediction in which a plurality of reference frames can be used.

  There are three types of image compression methods: I picture, P picture, and B picture. In the I (Intra) picture, other frames are not referred to. In the P picture, a past frame in time can be referred to, and in the B picture, a plurality of frames can be referred to.

  In a B picture, a plurality of reference frames can be used as a reference for motion. However, since the temporal position and number of reference frames can be changed for each macro block which is a motion compensation unit, depending on the macro block, You can refer to only one sheet instead of two.

  In the conventional motion vector search, considering the possibility of selecting unidirectional prediction referring to only one reference frame, the motion vector combination that minimizes the error for each reference frame is selected as the motion of bidirectional prediction. Adopted as a vector.

  However, even if the motion vector that minimizes the error with respect to the past reference frame and the motion vector that minimizes the error with respect to the future reference frame are obtained separately, the combination of these two motion vectors is the optimal bi-directional prediction. It does not determine the motion vector.

  Since bi-directional prediction uses the pixel average of the region derived by two motion vectors, there are other optimal motion vectors. Considering pixel averaging in multiple reference frame areas, even if the error for each reference frame is minimal, the overall error is not necessarily minimized. There is a problem.

  The moving picture coding method of Patent Document 1 generates a motion vector for bidirectional prediction by temporally dividing a motion vector between frames serving as reference frames, and a forward direction that minimizes an error. The motion vector and the backward motion vector are calculated separately, and they are still combined.

  Although the motion vector search method of Patent Document 2 requires a small amount of computation, it obtains a backward motion vector and uses this to find a forward motion vector. The forward motion vector and the backward motion vector Since an error regarding the combination is not taken into consideration, a combination of a forward motion vector and a backward motion vector that minimizes the error cannot be obtained.

  The video image encoding method of Patent Document 3 obtains a forward motion vector and a backward motion vector for a B frame by internally dividing a forward motion vector in time, and this is the order that minimizes the error. A combination of a directional motion vector and a backward motion vector cannot be obtained.

  In the moving image processing method described in Patent Literature 4, various adjustment values are added to the backward motion vector to calculate backward motion vector candidates, and the average value of the prediction data in both directions is the coding block. The one showing the highest degree of correlation is selected as the backward motion vector. However, what is once obtained for the forward motion vector is fixed, and the backward motion vector is selected under that condition, so that various combinations of other forward motion vectors and backward motion vectors are verified. Even if there are other combinations of forward motion vectors and backward motion vectors in which the error between the encoded block and the average value of the prediction data in both directions becomes smaller, the combination is not selected.

  An object of the present invention is to solve the above-described problems and provide an image encoding apparatus capable of performing encoding excellent in terms of compression efficiency and image quality in bidirectional prediction or bi-prediction.

  In order to solve the above problems, the present invention reconstructs an image signal from encoded information, generates a prediction signal based on the reconstructed image signals, and uses the prediction signal to generate an image signal to be encoded. An image encoding apparatus that predictively encodes and outputs encoded information, a difference unit that obtains a difference between an image signal to be encoded and a prediction signal, a transform unit that orthogonally transforms an output of the difference unit, A quantizing means for quantizing the output of the converting means; a variable length encoding means for variable length encoding the output of the quantizing means; an inverse quantization means for dequantizing the output of the quantizing means; Inverse transform means for inverse transforming the output of the inverse quantization means; Adder means for adding the output of the prediction signal and the output of the inverse transform means to output a reconstructed image signal; and the output of the adder means as a reference frame Frame memo temporarily stored as And an iterative motion search means for determining motion information for the plurality of reference frames by searching for motion information by an iterative process using an image signal to be encoded and a plurality of reference frames stored in the frame memory, It is characterized by comprising motion compensation means for outputting a prediction signal based on the plurality of reference frames subjected to motion compensation using motion information determined by the iterative motion search means.

  Also, the present invention is characterized in that the repetitive motion search means gradually derives motion information by sequentially using a plurality of reference frames in repetitive processing.

  In the present invention, the repetitive motion search means performs motion search while fixing motion information about reference frames other than the reference frame for motion search as the optimal motion information at that time, so that a plurality of reference frames can be obtained. It is characterized by optimizing a set of motion information.

  Further, according to the present invention, the repetitive motion search means includes a pixel value a multiplication unit for multiplying a pixel value of the first reference frame by a (a is a positive integer), and a pixel value of the second reference frame is b. In the first motion search, a motion search is performed on the image signal obtained by multiplying the pixel value of the image signal to be encoded by a times in the first motion search. In the motion search after the first time, a motion search unit that performs motion search on the image signal reflecting the previous motion compensation, and a motion compensation unit that performs motion compensation on the reference frame using the motion information searched by the motion search unit And a bias applying unit that preliminarily increases the pixel value of the image signal to be encoded by (a + b) times in the second and subsequent motion search by the iterative processing, and a code in which the pixel value is increased by (a + b) times by the bias applying unit. Image signal to be processed and reference compensated by the motion compensation unit A subtractor for obtaining a frame difference, and inputting the output of the subtractor to the motion search unit as an image signal reflecting the previous motion compensation, thereby weighting each of the first reference frame and the second reference frame. It is characterized by determining motion information in weighted prediction with a and b.

  Further, according to the present invention, the repetitive motion search means performs a motion search on the image signal itself to be encoded in the first motion search, and the previous motion compensation is reflected in the second and subsequent motion search by the iterative process. A motion search unit that performs motion search on the received image signal, a motion compensation unit that performs motion compensation on the reference frame using the motion information searched by the motion search unit, and a second or subsequent motion search by iterative processing. A bias applying unit that increases the pixel value of the image signal to be encoded by n times (n is an integer of 2 or more), the image signal to be encoded whose pixel value is increased by n times by the bias applying unit, and the motion compensation A subtraction unit that obtains a difference between reference frames that have been subjected to motion compensation and whose pixel value has been multiplied by (n−1) times, and outputs the output of the subtraction unit to the motion search unit as an image signal that reflects the previous motion compensation. By typing It is characterized by determining motion information for n reference frames.

  Further, the present invention is characterized in that the iterative motion search means includes means for terminating the iterative process based on the presence or absence of a change in motion information searched by the motion search unit or a predetermined number of iterations.

  Furthermore, the present invention is characterized in that the motion search unit can employ any search method.

  According to the present invention, motion search and motion compensation are repeated for bi-directional prediction and bi-prediction to obtain a motion vector for each reference frame, so that an optimal motion vector can be determined as a whole. it can. Thereby, it is possible to improve the compression efficiency in image encoding using a plurality of motion vectors.

  The present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of an image encoding device according to the present invention. The image coding apparatus of the present embodiment includes a subtractor 11, an orthogonal transformer (DCT) 12, a quantizer (Q) 13, a variable length coder (VLC) 14, an inverse quantizer (IQ) 15, and an inverse An orthogonal transformer (IDCT) 16, an adder 17, a frame memory (memory) 18, an iterative motion searcher (iterative ME) 19, and a motion compensator (MC) 20 are provided.

  The image signal (frame) to be encoded is input to the subtractor 11 and the iterative motion searcher 19. The subtracter 11 obtains an error between the image signal to be encoded and the prediction signal transmitted from the motion compensator 20. The error signal output from the subtractor 11 is orthogonally transformed by the orthogonal transformer 12 and becomes an orthogonal transformation coefficient. This orthogonal transform coefficient is quantized by the quantizer 13, further variable-length encoded by the variable-length encoder 14, and encoded and output as a bit stream.

  The output of the quantizer 13 is inversely quantized by the inverse quantizer 15, further inversely orthogonally transformed by the inverse orthogonal transformer 16, and input to the adder 17. The adder 17 adds the output obtained by the inverse orthogonal transform process and the prediction signal transmitted from the motion compensator 20, and outputs a reconstructed (locally decoded) image signal. The frame memory 18 temporarily stores the image signal reconstructed by the adder 17 as a reference frame.

  The iterative motion searcher 19 uses the encoding target frame and a plurality of reference frames stored in the frame memory 18 to determine motion information by searching for motion information by iterative processing. Thereby, the optimal combination of motion information in consideration of the entire plurality of reference frames is determined. Details of the processing here will be described later.

  The motion compensator 20 performs motion compensation on a plurality of reference frames using the motion information determined by the motion searcher 19, and outputs a prediction signal.

  FIG. 2 is a functional block diagram showing a first configuration example of the iterative motion searcher (iterative ME) 19. This iterative motion searcher 19 inputs a frame to be encoded and two reference frames (reference frames 1 and 2) and outputs a bi-predicted motion vector MV. Further, it is characterized in that the accuracy is gradually improved by repeating the unidirectional motion vector search repeatedly.

  The iterative motion searcher 19 includes a motion search unit 21, a motion compensation unit 22, a bias applying unit 23, a subtraction unit 24, a determination unit 25, and switching units 26 and 27.

  The switching unit 26 inputs the encoding target frame to the motion search unit 21 in the first first motion search, and inputs the output of the subtraction unit 24 to the motion search unit 21 in the second and subsequent motion searches.

  Therefore, the motion search unit 21 performs a motion search on the image signal itself to be encoded in the first motion search for the encoding target frame, and the previous motion compensation is reflected in the second and subsequent motion searches by iterative processing. A motion search is performed on the image signal.

  Further, the switching unit 27 alternately inputs the first reference frame 1 and the second reference frame 2 to the motion search unit 21 for each motion search. The motion compensation unit 22 performs motion compensation on the reference frame using the motion information searched by the motion search unit 21.

  The bias applying unit 23 biases the encoding target frame in advance for the second and subsequent motion search by the iterative process. This eliminates the influence of reference frames other than the reference frame for motion search.

  The subtracting unit 24 obtains a difference between the encoding target frame biased by the bias applying unit 23 and the reference frame motion-compensated by the motion compensating unit 22. The determination unit 25 ends the iterative process based on the presence or absence of a change in the motion information searched by the motion search unit 21 or a predetermined number of iterations.

  Hereinafter, the operation of the iterative motion searcher 19 will be described. In the first motion search, the switching units 26 and 27 are switched to input the encoding target frame and the first reference frame 1 to the motion search unit 21. The motion search unit 21 performs motion search on the encoding target frame itself at the first time. That is, the motion vector MV is searched using the encoding target frame and the reference frame 1.

  The determination unit 25 determines whether or not there is a change in the motion vector MV and the motion search is within the specified number of times. As a result of the first motion search, the motion vector MV changes and is determined to be within the specified number of times, so that the searched motion vector MV is given to the motion compensation unit 22. The motion compensation unit 22 generates a prediction frame using the motion vector MV searched by the motion search unit 21 and the reference frame 1.

  The determination unit 25 is based on whether or not it is within the specified number of times in consideration of the case where the motion vector MV does not converge no matter how many times the motion search is repeated. It is set to a predetermined number that exceeds twice the number. Note that the determination of whether or not there is a change in the motion vector MV does not have to be whether or not the change in the motion vector MV is completely 0, but whether or not it is equal to or less than a preset threshold value. But you can.

  On the other hand, the encoding target frame is input to the bias applying unit 23, and the pixel value is doubled. As a result, the subtraction unit 24 obtains the difference between the encoding target frame whose pixel value has been doubled and the prediction frame created by the motion compensation unit 22.

  In the second motion search, a motion search is performed on the image signal reflecting the first motion compensation. That is, motion search is performed using the difference output from the subtractor 24 instead of the encoding target frame. The second reference frame 2 is used as the reference frame. Therefore, the switching units 26 and 27 are switched so that the difference from the subtraction unit 24 and the reference frame 2 are input to the motion search unit 21.

  The pixel value of the encoding target frame is doubled by the bias applying means 23, and the subtraction unit 24 outputs the encoding target frame in a form in which the residual signal from the previous motion compensation remains, so two frames are output. The same motion search unit 21 can be used in the motion search using the eye reference frame 2.

  In the second motion search, the motion search unit 21 searches for the motion vector MV using the difference from the subtraction unit 24 and the reference frame 2. Since the motion vector MV for the reference frame 2 has been searched for the first time this time, the determination unit 25 determines that the motion vector MV also changes this time and is within the specified number of times. Therefore, the searched motion vector MV is given to the motion compensation unit 22.

  The motion compensation unit 22 generates a prediction frame using the motion vector MV searched by the motion search unit 21 and the reference frame 2, and the subtraction unit 24 encodes the pixel value whose pixel value is doubled by the bias applying unit 23 The difference between the frame and the predicted frame created by the motion compensation unit 22 is output.

  In the third motion search, the difference output from the subtraction unit 24 for the second time is used. The first reference frame 1 is used as the reference frame. That is, the switching unit 26 is in the same state as the second time, and the switching unit 27 is switched to input the reference frame 1 to the motion search unit 21.

  Thereafter, similarly, a series of processes of motion search and motion compensation are repeated repeatedly. As described above, the iterative motion searcher 19 performs a motion search by using the encoding target frame itself and the reference frame 1 at the first time, and after the second time, the image signal that reflects the previous motion compensation, that is, The motion search is repeated by using the difference from the subtracting unit 24 and alternately using the reference frames 1 and 2. By switching the reference frame used in the motion search unit 21 and performing iterative processing, the overall optimal motion vector MV can be obtained gradually.

  If the calculated motion vector MV no longer changes during the motion search iteration process, or if the number of iterations of a series of processes exceeds a predetermined number of times, the last iteration process is performed for reference frames 1 and 2. The obtained motion vector MV is output and the motion search process is terminated.

  The motion compensator 20 (FIG. 1) generates a prediction frame for the encoding target frame using the motion vector MV for the reference frames 1 and 2 obtained as a result of the search and the image signals of the reference frames 1 and 2. The generation of the prediction frame is the same as the conventional bi-prediction and bi-prediction, and thus description thereof is omitted.

  FIG. 3 is an explanatory diagram of motion search in the first configuration example using specific samples. Here, in order to facilitate the understanding of the operation of the iterative motion searcher 19, it is simplified and the frame is composed of 2 × 1 pixels, and the motion search is performed in units of 1 × 1 pixels. However, in practice, the number of pixels in the frame and the unit of motion search are 352 × 288 pixels, 16 × 16 pixels, and the like, respectively.

   Assuming that the pixel value to be encoded is “6” and the pixel values of reference frames 1 and 2 are “4, 5” and “5, 8” respectively, an error will occur for reference frames 1 and 2. , The position of the pixel value “5” at which the error is minimized from the reference frame 1 is determined as the motion vector with respect to the pixel value “6” to be encoded. From the reference frame 2, the position of the pixel value “5” is determined as a motion vector. In this case, the pixel value obtained by averaging in bidirectional prediction is “5”, and the error with respect to the pixel value “6” to be encoded is “1”.

  On the other hand, when confirming the operation against the processing of the iterative motion searcher 19, first, the motion search unit 21 uses the pixel value “5” of the reference frame 1 that minimizes the error with respect to the pixel value “6” to be encoded. Is obtained as a motion vector. Up to this point, the same as above.

  Here, since the motion vector for the reference frame 1 is obtained for the first time, the determination unit 25 gives the motion vector to the motion compensation unit 22 on the assumption that the condition is satisfied. Next, the pixel value “5” of the reference frame 1 is associated with the pixel value “6” to be encoded as the pixel value of the prediction pixel by the motion compensation in the motion compensation unit 22.

  On the other hand, the pixel value “6” to be encoded is preliminarily doubled to “12” by the bias applying unit (pixel value doubling means) 23. In the subtraction unit 24, a difference “7” between the value “12” and the pixel value “5” subjected to motion compensation by the motion compensation unit 22 is obtained.

  In the second motion search, the motion search unit 21 searches for the position of the pixel value of the reference frame 2 that minimizes the difference from the difference “7” as a motion vector. In this case, the position of the pixel value “8” of the reference frame 2 is obtained as a motion vector. Since the motion vector for the reference frame 2 is obtained for the first time, the determination unit 25 gives the motion vector to the motion compensation unit 22 assuming that the condition is satisfied. Next, by the motion compensation in the motion compensation unit 22, the pixel value “8” of the reference frame 2 is associated with the pixel value “6” to be encoded as the pixel value of the prediction pixel.

  The subtracting unit 24 obtains a difference “4” between the pixel value “12” to be encoded that has been previously biased by the bias applying unit 23 and the pixel value “8” that has been motion compensated by the motion compensating unit 22.

  In the third motion search, the motion search unit 21 searches for the position of the pixel value of the reference frame 1 that minimizes the difference from the difference “4” as a motion vector. In this case, the position of the pixel value “4” of the reference frame 1 is obtained as a motion vector. Here, since the motion vector for the reference frame 1 has changed, the determination unit 25 gives the motion vector to the motion compensation unit 22 assuming that the condition is satisfied. Next, the pixel value “4” of the reference frame 1 is associated with the pixel value “6” to be encoded as the pixel value of the prediction pixel by motion compensation in the motion compensation unit 22.

  Next, the subtraction unit 24 obtains a difference “8” between the value “12” obtained from the bias applying unit 23 and the pixel value “4” motion-compensated by the motion compensation unit 22.

  In the fourth motion search, the motion search unit 21 searches for the position of the pixel value of the reference frame 2 that minimizes the difference from the difference “8” as a motion vector. Here, the position of the pixel value “8” in the reference frame 2 is obtained as a motion vector, and the motion vector for the reference frame 2 remains at the position of the pixel value “8” and does not change. The determination unit 25 determines that the condition is not satisfied, finishes the series of iterative processes, and finally determines the position of the pixel value “4” in the reference frame 1 and the position of the pixel value “8” in the reference frame 2 Output as a motion vector.

  In this case, the average pixel value obtained by bi-directional prediction is “6”, and the error for the encoding target pixel value “6” is “0”. It can be seen that the optimal motion vector is selected.

  FIG. 4 is a functional block diagram showing a second configuration example of the iterative motion searcher (iterative ME) 19. This configuration example is generalized so as to be compatible with unidirectional or bidirectional weighted prediction using a plurality of reference frames. In FIG. 4, the same or equivalent parts as in FIG.

  The second configuration example is different from the first configuration example in that the bias applying unit 23 is a pixel value (a + b) multiplication unit, the encoding target frame is provided to the switching unit 27 through the pixel value a multiplication unit 29, and the reference frame 1 and 2 are given to the switching unit 27 through the pixel value a multiplication unit 30 and the pixel value b multiplication unit 31, respectively. Here, a and b indicate weights for the reference frames 1 and 2, respectively, and a = b = 1 corresponds to the first configuration example.

  FIG. 5 is an explanatory diagram of motion search in the second configuration example using specific samples. Here, it is assumed that a = 2, b = 1, the pixel value to be encoded is “5”, and the pixel values of the reference frames 1 and 2 are “1, 3” and “12, 13”, respectively.

  When the motion vector that minimizes the error for each of the reference frames 1 and 2 is obtained individually, the error taking into account the weights a and b is different from the reference frame 1 and 2 for the pixel value “5” to be encoded. The positions of the minimum pixel values “3” and “12” are determined as motion vectors. In this case, the pixel value obtained by weighted averaging in bidirectional prediction is “6”, and the error with respect to the pixel value “5” to be encoded is “1”.

  On the other hand, when the operation is confirmed in light of the process of the iterative motion searcher 19, first, the motion search unit 21 refers to the value “10” obtained by doubling the pixel value to be encoded by the pixel value a multiplication unit 29. An error from the value “2, 6” obtained by doubling the pixel value of frame 1 by the pixel value a multiplication unit 29 is obtained, and the position of the pixel value of reference frame 1 at which this error is minimized is searched for as a motion vector. Here, the position of the pixel value “3” of the reference frame 1 is obtained as a motion vector. This is the same as above.

  Since the motion vector for the reference frame 1 is obtained for the first time, the determination unit 25 gives the motion vector to the motion compensation unit 22 assuming that the condition is satisfied. Next, by motion compensation in the motion compensation unit 22, a value “6” that is twice the pixel value of the reference frame 1 is associated with the pixel value “5” to be encoded as the pixel value of the prediction pixel.

  On the other hand, the pixel value to be encoded is preliminarily tripled to “15” by the bias applying unit (pixel value (a + b) multiplication means) 23. The subtractor 24 obtains a difference “9” between the value “15” and the pixel value “6” motion-compensated by the motion compensator 22.

  In the second motion search, the motion search unit 21 obtains an error between the difference “9” and the value “12, 13” obtained by multiplying the pixel value of the reference frame 2 by the pixel value b multiplication unit 30. The position of the pixel value of the reference frame 2 that minimizes is searched as a motion vector. In this case, the position of the pixel value “12” of the reference frame 2 is obtained as a motion vector. Since the motion vector for the reference frame 2 is obtained for the first time, the determination unit 25 gives the motion vector to the motion compensation unit 22 assuming that the condition is satisfied. Next, by motion compensation in the motion compensation unit 22, a value “12” that is one time the pixel value of the reference frame 2 is associated with the pixel value “5” to be encoded as the pixel value of the prediction pixel.

  The subtracting unit 24 obtains a difference “3” between the pixel value “15” to be encoded that has been previously biased by the bias applying unit 23 and the pixel value “12” that has been motion compensated by the motion compensating unit 22.

  In the third motion search, the motion search unit 21 refers to a reference that minimizes an error between the difference “3” and the value “2, 6” obtained by doubling the pixel value of the reference frame 1 by the pixel value a multiplication unit 29. The position of the pixel value of frame 1 is searched as a motion vector. In this case, the position of the pixel value “1” of the reference frame 1 is obtained as a motion vector. Here, since the motion vector for the reference frame 1 has changed, the determination unit 25 gives the motion vector to the motion compensation unit 22 assuming that the condition is satisfied. Next, by motion compensation in the motion compensation unit 22, a value “2” that is twice the pixel value of the reference frame 1 is associated with the pixel value “5” to be encoded as the pixel value of the prediction pixel.

  Next, the subtraction unit 24 obtains a difference “13” between the value “15” obtained from the bias applying unit 23 and the pixel value “2” motion-compensated by the motion compensation unit 22.

  In the fourth motion search, the motion search unit 21 obtains an error between the difference “13” and the value “12,13” obtained by multiplying the pixel value of the reference frame 2 by the pixel value b multiplication unit 30. The position of the pixel value of the reference frame 2 that minimizes is searched as a motion vector. In this case, the position of the pixel value “13” of the reference frame 2 is obtained as a motion vector. Here, since the motion vector for the reference frame 2 has changed, the determination unit 25 gives the motion vector to the motion compensation unit 22 assuming that the condition is satisfied. Next, by motion compensation in the motion compensation unit 22, a value “13” that is one time the pixel value of the reference frame 2 is associated with the pixel value “5” to be encoded as the pixel value of the prediction pixel.

  Next, the subtraction unit 24 obtains a difference “2” between the value “15” obtained from the bias applying unit 23 and the pixel value “13” motion-compensated by the motion compensation unit 22.

  In the fifth motion search, the motion search unit 21 calculates an error between the difference “2” and the value “2, 6” obtained by doubling the pixel value of the reference frame 1 by the pixel value a multiplication unit 29. The position of the pixel value of the reference frame 1 that minimizes is searched as a motion vector. In this case, the position of the pixel value “1” of the reference frame 1 is obtained as a motion vector. Here, the motion vector for the reference frame 1 remains at the position of the pixel value “1” and does not change. The determination unit 25 determines that the condition is not satisfied, ends the above series of iterative processes, and finally determines the position of the pixel value “1” in the reference frame 1 and the position of the pixel value “13” in the reference frame 2 Output as a motion vector.

  In this case, the pixel value obtained by weighted averaging in bi-directional prediction is “5”, and the error with respect to the pixel value “5” to be encoded is “0”. An optimal motion vector is selected as a whole.

  FIG. 6 is a functional block diagram showing a third configuration example of the iterative motion searcher (iterative ME) 19. In FIG. 6, the same or equivalent parts as in FIG. In the first configuration example, it is assumed that there are two reference frames. However, in the third configuration example, it is possible to cope with an arbitrary number (n) of three or more reference frames. The third configuration example is different from the first configuration example in that the bias applying unit 23 is a pixel value n-times multiplying unit, and the pixel value of the prediction frame generated by the motion compensation unit 22 is a pixel value (n-1) multiplication unit 28. This is a point that is multiplied by (n-1) and given to the subtracting unit 24.

  Also in this case, the subtracter 24 outputs the encoding target frame in which the residual signal in the previous motion compensation remains as a difference, so that the same motion is detected in the next motion search as in the first specific example. The search unit 21 can be used.

  As mentioned above, although embodiment was described, this invention is not limited to the said embodiment. For example, the motion search unit 21 can employ any search method, and can appropriately select and apply an appropriate search method.

It is a block diagram which shows one Embodiment of the image coding apparatus which concerns on this invention. It is a functional block diagram which shows the 1st structural example of an iterative motion searcher. It is explanatory drawing of the motion search in the 1st structural example by a specific sample. It is a functional block diagram which shows the 2nd structural example of an iterative motion searcher. It is explanatory drawing of the motion search in the 2nd structural example by a specific sample. It is a functional block diagram which shows the 3rd structural example of an iterative motion search device.

Explanation of symbols

11 ... Subtractor, 12 ... Orthogonal transformer (DCT), 13 ... Quantizer (Q), 14 ... Variable length encoder (VLC), 15 ... Inverse quantizer (IQ), 16 ... Inverse orthogonal transformer (IDCT), 17 ... Adder, 18 ... Frame memory (memory), 19 ... Repetitive motion searcher (iterative ME), 20 ... Motion compensator (MC), 21 ... motion search unit, 22 ... motion compensation unit, 23 ... bias applying unit, 24 ... subtraction unit, 25 ... determination unit, 26, 27 ...・ Switching unit, 28 ... Pixel value (n-1) multiplication unit, 29,30 ... Pixel value a multiplication unit, 31 ... Pixel value b multiplication unit

Claims (7)

An image that reconstructs an image signal from encoded information, generates a prediction signal based on the reconstructed plurality of image signals, predictively encodes an image signal to be encoded using the prediction signal, and outputs encoded information An encoding device comprising:
A difference means for obtaining a difference between the image signal to be encoded and the prediction signal;
Transform means for orthogonally transforming the output of the difference means;
Quantization means for quantizing the output of the conversion means;
Variable length encoding means for variable length encoding the output of the quantization means;
Inverse quantization means for inversely quantizing the output of the quantization means;
Inverse transform means for inverse transforming the output of the inverse quantization means;
Adding means for adding the output of the prediction signal and the inverse transform means to output a reconstructed image signal;
A frame memory for temporarily storing the output of the adding means as a reference frame;
Repetitive motion search means for determining motion information for the plurality of reference frames by searching for motion information by iterative processing using an image signal to be encoded and a plurality of reference frames stored in the frame memory;
An image encoding apparatus comprising: motion compensation means for outputting a prediction signal based on the plurality of reference frames subjected to motion compensation using motion information determined by the repetitive motion search means. The image coding apparatus according to claim 1, wherein the repetitive motion search means gradually derives motion information by sequentially using a plurality of reference frames in repetitive processing. The repetitive motion search means performs a motion search while fixing motion information for reference frames other than the reference frame for motion search as the optimal motion information at that time, thereby providing a set of motion information for a plurality of reference frames. The image coding apparatus according to claim 2, wherein the image coding apparatus is optimized. The repetitive motion search means includes a pixel value a multiplication unit for multiplying a pixel value of the first reference frame by a (a is a positive integer) and a pixel value of the second reference frame by b (b is positive). In the first motion search, a motion search is performed on the image signal obtained by multiplying the pixel value of the image signal to be encoded by a times, and in the second and subsequent motion search by iterative processing. A motion search unit that performs a motion search on an image signal that reflects the previous motion compensation, a motion compensation unit that performs motion compensation on a reference frame using the motion information searched by the motion search unit, and two by iterative processing. A bias applying unit for multiplying a pixel value of an image signal to be encoded by (a + b) times in advance for a motion search after the first time, and an image signal to be encoded whose pixel value has been increased by (a + b) times by the bias applying unit; The difference between the reference frames motion compensated by the motion compensation unit is obtained. A subtracting unit that inputs the output of the subtracting unit to the motion search unit as an image signal reflecting the previous motion compensation, whereby weights a and b are respectively given to the first reference frame and the second reference frame. 4. The image coding apparatus according to claim 3, wherein motion information in the attached weighted prediction is determined. The repetitive motion search means performs a motion search on the image signal itself to be encoded in the first motion search, and on an image signal reflecting the previous motion compensation in the second and subsequent motion search by the iterative process. A motion search unit that performs motion search, a motion compensation unit that performs motion compensation of a reference frame using the motion information searched by the motion search unit, and an image signal to be encoded in advance in a second or subsequent motion search by iterative processing A bias applying unit for multiplying the pixel value of n by n (n is an integer of 2 or more), an image signal to be encoded whose pixel value has been multiplied by n by the bias applying unit, and motion compensation by the motion compensation unit, A subtractor that obtains a difference between reference frames whose pixel values have been multiplied by (n-1) times, and the output of the subtractor is input to the motion search unit as an image signal in which the previous motion compensation is reflected. Pieces of reference frame The image encoding apparatus according to claim 3, wherein motion information about the video is determined. 6. The repetitive motion search means, comprising: means for terminating repetitive processing based on the presence or absence of a change in motion information searched by the motion search unit or a predetermined number of repetitions. Image encoding device. The image coding apparatus according to claim 4, wherein the motion search unit can employ any search method.

Images