JPH07240927A - Video encoder - Google Patents

Abstract

PURPOSE:To provide the video encoder which can reduce circuit scale in comparison with the case of using a full searching method and can be provided with picture quality close to the full search method. CONSTITUTION:A search information calculator 102 calculates one candidate vector from motion vector values stored in a motion vector memory 101. A high-accuracy computing element 103 sets a second search area which includes a block to be predicted by the candidate vector and is smaller than a search area on a reference frame, sets a candidate vector with accuracy smaller than one picture element and calculates a predictive error value for each candidate vector. A low-accuracy computing element 104 sets a candidate vector with accuracy larger than two picture elements inside the first search area on the reference frame and calculates a predictive error value for each candidate vector. A predictive error comparator 105 compares the predictive error values calculated by the high-accuracy and low-accuracy computing elements 103 and 104 so that the motion vector value of the encoded block can be decided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a video coding apparatus which performs high efficiency coding using motion compensation in order to reduce the data amount of a video signal when recording and transmitting a video.

[0002]

2. Description of the Related Art In recent years, development of a device which digitally encodes a video signal has become popular due to advances in digital signal processing technology of video and semiconductor integration technology. One of the video signal encoding devices is one that reduces redundancy in the time direction. As a typical example of this, so-called MPEG (Mo
ving Picture Expert Group), and standardization work is currently underway. In MPEG, DCT (Discrete Cosine Transform) is used for compression in the spatial direction, and motion-compensated interframe prediction is used for compression in the temporal direction.

Motion-compensated inter-frame prediction refers to an input video signal having a certain size (for example, 16 horizontal pixels and 16 vertical pixels).
It is divided into blocks, and the prediction is performed by shifting the motion vector from the reference frame which is the reproduced video signal of the previous frame, and the video signal is encoded by the difference signal between the input signal and the prediction signal and the motion vector. .

The motion vector detection is performed by matching the coded block and a block on the reference frame corresponding to the position of the coded block as a reference position with a block group in a preset search area, This is a process of detecting the position of a block having a high correlation between blocks as a motion vector.

Here, a conventional video encoding apparatus will be described with reference to the drawings. FIG. 15 is a block diagram showing the overall configuration of a conventional video encoding device. In the figure, a buffer memory 1501 stores video data. A blocker 1502 divides the video data stored in the buffer memory 1501 into a plurality of blocks. A frame memory 1503 holds the video data of the previous frame. A motion vector detector 1504 detects a motion vector from the coded block by the blocker 1502 and the video data held in the frame memory 1403 using search information. A motion vector encoder 1505 encodes the motion vector detected by the motion vector detector 1504. A prediction block generator 1506 generates, as a prediction block, video data of a block displaced by the motion vector from the video data held in the frame memory 1503. 1507
Is a differencer, which calculates the difference signal between the coding block and the prediction block. Reference numeral 1508 denotes an orthogonal transformer, which orthogonally transforms video data. A quantizer 1509 quantizes the orthogonally transformed video data. An orthogonal transform encoder 1510 encodes the orthogonally transformed and quantized video data. Reference numeral 1511 denotes an inverse quantizer, which performs the inverse operation of the quantizer. An inverse orthogonal transformer 1512 decodes the orthogonally transformed video data. An adder 1513 adds the decoded video data and the prediction block. 1514, 1515
Is a switch that switches the input signal depending on the presence or absence of the prediction block.

The operation of the conventional video coding apparatus thus configured will be described. First, the video data of the first frame is stored in the buffer memory 1501. The stored video data is supplied to the blocker 1502, and the screen of the first frame is divided into a plurality of blocks so as to form a checkerboard pattern. For example, one block has 16 pixels vertically and 16 pixels horizontally. Since the predictive video data does not exist in the video data of the first frame, the switch 1514 is connected to the orthogonal transformer 1508, and the video data of each block is the orthogonal transformer 150.
8 and is subjected to orthogonal transformation. The transform coefficient is quantized by a quantizer 1509, and an orthogonal transform encoder 1510 is used.
Is encoded by.

On the other hand, the quantized data of the quantizer 1509 is also given to the inverse quantizer 1511 and subjected to inverse quantization processing. Further, the inverse orthogonal transformer 1512 converts the original video data again. At this time, the switch 1515 causes the adder 1513 to be in a through state, and no prediction block is given. One frame of video data output by the adder 115 is stored in the frame memory 1503 and is used for detecting the motion vector of the next frame.

Now, when the video data of the next frame is output from the signal source, the data is temporarily stored in the buffer memory 1.
It is stored in 501. The stored video data is divided into a plurality of blocks by the blocker 1502. In each block, each video data is searched within a predetermined search range by the motion vector detector 1504, with the position corresponding to the video data one frame before as a reference point. For example, the search range is set to 7 pixels in each of the left and right and up and down directions, and the prediction error between the adjacent frames (signal level difference)
Is determined as a motion vector.

Next, the prediction block generator 1506 uses the motion vector data obtained by the motion vector detector 1504 to predict the image displaced by the motion vector from the reference position one frame before, and predicts the prediction block. To generate.
The prediction block generated here is given to the differentiator 1507 and converted into difference data with respect to the corresponding block in the previous frame.

The differential data obtained by the differentiator 1507 is given to the orthogonal transformer 1508 via the switch 1514, and orthogonal transform for removing spatial redundancy of the signal is performed. The transform coefficient here is quantized by the quantizer 1509, and encoded by the orthogonal transform encoder 112. On the other hand, the motion vector detected by the motion vector detector 1504 is encoded using the motion vector encoder 1505.

Also, the quantized data is the inverse quantizer 1511.
Is inversely processed by the quantizer 1509, and is inversely transformed into differential data by the inverse orthogonal transformer 1512. Then, in the adder 1513, the prediction block of the video given via the switch 1515 is added to the difference data to be returned to the original video data. The returned video data is stored in the frame memory 1503 and used for detecting the motion vector of the next block.

[0012] In addition to the video data of the first frame, the code amount may be smaller when the input signal itself of the block to be encoded is converted and encoded than when the difference data is converted. . For example, when the entire screen becomes a new video, it is necessary to encode the input data as it is. At this time, the output of the blocker 1502 is input to the orthogonal transformer 1508 using the switch 1514 so that the prediction data is not given to the adder 1513.

As a typical motion vector detection method,
There are a full search method, a three-step method, and the like. The full search method is a method of searching for a motion vector with 1-pixel accuracy over the entire search area, and is the best method in terms of image quality, but has a problem of large circuit scale. The three-step method is a method devised to solve the circuit scale problem of the full search method, and detects a motion vector by three-stage processing. That is, first, in the first step, a motion vector is detected by searching with a precision of 4 pixels. Next, in the second step, the motion vector detected in the first step and its eight neighbors are searched with a 2-pixel precision to detect the motion vector. Furthermore, in the third step, the motion vector of the coding block is determined by searching the motion vector detected in the second step and its eight neighbors with one pixel precision. In the 3-step method, the number of searches is smaller than in the full search method, and the circuit size can be reduced. However, in terms of image quality, deterioration is seen compared to the full search method.

FIG. 16 shows the search area on the coding block and the reference frame when the search range of the motion vector is set from -7 to +7 in the horizontal direction and from -7 to +7 in the vertical direction. Here, the size of the coding block is 16 in the horizontal direction.
There are 16 pixels in the vertical direction. The size of the search area is 30 pixels in the horizontal direction and 30 pixels in the vertical direction.

FIG. 17 shows the number of motion vector searches when the full search method is used. When the search range of the motion vector is −7 to +7 in the horizontal direction and −7 to +7 in the vertical direction, the number of searches is 225.

FIG. 18 shows the number of motion vector searches when the three-step method is used. As shown in the figure, the first
The number of searches is 9 in the step, the number of searches is 8 in the second step, the number of searches is 8 in the third step, and the total number of searches is 2
It becomes 5.

[0017]

In such a conventional video coding apparatus, when the full search method is used, the number of motion vector searches is large and the circuit implementation becomes large. Further, when the three-step method is used, there is a problem that the image quality is deteriorated.

In view of the above problems of the conventional video coding apparatus, the present invention can reduce the circuit scale as compared with the case of using the full search method, and has an image quality close to that of the full search method. It is an object of the present invention to provide a video encoding device capable of having the following.

[0019]

In order to achieve the above object, the video coding apparatus of the present invention uses a search area on a preset reference frame for detecting a motion vector as a first search area. Then, one candidate vector is calculated from the motion vector memory that stores the motion vector value of the coded block in which the motion vector is detected, and the motion vector value stored in the motion vector memory by a predetermined arithmetic expression. A candidate vector calculator and a second search area including a block predicted by the candidate vector and smaller than the search area on the reference frame are set, and the candidate vector is set in the second search area with an accuracy of 1 pixel or less. And a high-precision calculator that calculates a prediction error value for each candidate vector and a candidate vector with an accuracy of 2 pixels or more in the first search area on the reference frame. By setting the prediction error value for each candidate vector and comparing the prediction error value calculated by the high-precision calculator with the prediction error value calculated by the high-precision calculator It is provided with a prediction error value comparator for determining the motion vector value of the coding block.

[0020]

With the above-described structure, the video coding apparatus of the present invention first uses the search information calculator to calculate 1 from the motion vector value stored in the motion vector memory.
Calculate candidate vectors. Next, the high-precision arithmetic unit sets a candidate vector with an accuracy of 1 pixel or less including a block predicted by the candidate vector, and calculates a prediction error value for each candidate vector. The low-precision calculator sets a candidate vector in the first search area on the reference frame with an accuracy of 2 pixels or more, and calculates a prediction error value for each candidate vector. Finally, the prediction error value comparator determines the motion vector value of the coding block by comparing the prediction error value calculated by the high precision arithmetic unit with the prediction error value calculated by the low precision arithmetic unit.

[0021]

【Example】

(Embodiment 1) A video encoding apparatus according to a first embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of a video encoding apparatus according to the first embodiment of the present invention. Here, the motion vector detector 15 of the video encoding device in the conventional example shown in FIG. 15 is used.
Since 04 is different, those configurations are described.

In FIG. 1, a search area on a preset reference frame for detecting a motion vector is referred to as a first search area. A motion vector memory 101 stores a motion vector value of a coded block in which a motion vector is detected. A search information calculator 102 calculates one candidate vector from a motion vector value stored in the motion vector memory 101 by a predetermined arithmetic expression. Reference numeral 103 denotes a high-precision arithmetic unit that sets a second search area including a block predicted by the candidate vector and smaller than the search area on the reference frame, and sets the candidate vector in the second search area with an accuracy of 1 pixel or less. After setting, a prediction error value is calculated for each of the candidate vectors. A low-precision calculator 104 sets a candidate vector in the first search area on the reference frame with an accuracy of 2 pixels or more, and calculates a prediction error value for each candidate vector. A prediction error value comparator 105 compares the prediction error value calculated by the high-precision calculator 103 with the prediction error value calculated by the low-precision calculator 104 to determine the motion vector value of the coding block. To do.

FIG. 2 is a flow chart for explaining the operation of the detailed block shown in FIG. Below, the processing procedure is shown for the video data of the 1st frame after the 2nd frame. (Step 201) The number of pixels in the frame is 800 horizontal pixels and 800 vertical pixels, the block size is 16 horizontal pixels and 16 vertical pixels, the total number of horizontal blocks is 50, and the total number of vertical blocks is 50. The initial value is 0 and the vertical block initial value is 0. Next, go to (step 202). (Step 202) The motion vector memory 101 is initialized to 0. Next, go to (step 203). (Step 203) Use the search information calculator 102 to calculate the candidate vector of the coding block. The candidate vector value in the horizontal direction is HP, and the candidate vector value in the vertical direction is VP
And (Step 204) and (Step 205)
Go to (Step 204) The high-precision arithmetic unit 103 detects a motion vector by using the data of the encoded block and the data of the search area by the method shown in the flowchart of FIG.
Assuming that 8 pixels adjacent to the candidate vector are searched, the search range of the horizontal motion vector is HP-1 to HP-1.
HP + 1, the search range of the vertical motion vector is VP−
1 to + VP + 1, the horizontal search accuracy is 1 pixel, and the vertical search accuracy is 1 pixel. Further, the minimum prediction error value and the initial setting value of its motion vector are not given. Here, the detected motion vector is HH in the horizontal direction and VH in the vertical direction, and the minimum prediction error value at that time is EH. Next, go to (step 206). (Step 205) The low-precision arithmetic unit 104 detects the motion vector using the data of the encoded block and the data of the search area by the method shown in the flowchart of FIG.
At this time, the search range of the horizontal motion vector is -4 to
+4, vertical motion vector search range is -4 to +
4. The search accuracy in the horizontal direction is 4 pixels, and the search accuracy in the vertical direction is 4 pixels. Also, the minimum prediction error value and the initial setting value of its motion vector are not given. Here, the detected motion vector is HL in the horizontal direction and VL in the vertical direction, and the minimum prediction error value at that time is EL. Next (Step 20
Go to 6). (Step 206) Using the prediction error value comparator 105,
The prediction error value EH of the motion vector detected by the high-precision calculator 103 is compared with the prediction error value EL of the motion vector detected by the low-precision calculator 104. If EL is smaller than EH, HL and VL are If not, HH and VH are determined as motion vector values of the coding block. Next, go to (step 209). (Step 207) The motion vector determined by the prediction error value comparator 105 is stored in the motion vector memory 101.
Next, go to (step 208). (Step 208) 1 in horizontal block of encoded frame
Is added. Next, go to (step 209). (Step 209) It is checked whether the horizontal block of the encoded frame is equal to 50, which is the total number of horizontal blocks. If they are equal, go to (step 210). If not,
Go to (step 203). (Step 210) The horizontal block of the encoded frame is set to 0.
Reset to. Add 1 to the vertical block of the encoded block. Next, go to (step 211). (Step 212) It is checked whether the vertical block of the encoded frame is equal to 50, which is the total number of vertical blocks. If they are equal, the process ends. If not, go to (step 203).

Here, after the motion vector is detected by each of the high-precision arithmetic unit 103 and the low-precision arithmetic unit 104, the prediction error value comparator 105 compares the prediction error values with each other, and Although the motion vector is determined, the high-precision arithmetic unit 103 extracts all prediction error values between the encoded block and the block corresponding to each search point, and the low-precision arithmetic unit 104 extracts the encoded block and each search. By extracting all prediction error values with the block corresponding to the point and comparing all those prediction error values in the prediction error value comparator 105,
The motion vector of the coding block may be determined.

FIG. 3 is a flowchart of the motion vector detector. The processing procedure is shown below. (Step 301) Horizontal motion vector search range is H_SRT to H_END, vertical motion vector search range is V_SRT to V_END, horizontal search accuracy is H_INTERVAL, and vertical search accuracy is V_.
Set to INTERVAL. Next (step 302)
Go to (Step 302) Initialize the minimum value of the prediction error value and its motion vector. If an initial value is given, set that value. When the initial value is not given, a value larger than the maximum value that can be taken as the minimum prediction error value is set, and the initial value of the motion vector is not set. Next (Step 30
Go to 3). (Step 303) Set the horizontal motion vector to H_SRT,
Set the vertical motion vector to V_SRT. Next, go to (step 304). (Step 304) It is checked whether the motion vector value has been initialized. If so, then go to (step 310). Otherwise (Go to step 305). (Step 305) A reference block, which is a block at a position shifted by the motion vector from the corresponding position of the encoded block in the search area, is created. Next, go to (step 306). (Step 306) A prediction error value between the coded block and the reference block is obtained. Here, as the method of calculating the prediction error value, the sum of absolute values of the difference values of corresponding pixels between blocks is used. Next, go to (step 307). (Step 307) The calculated prediction error value and the minimum prediction error value are compared. When the calculated prediction error value is smaller than the minimum prediction error value, the procedure goes to (step 308). If not, go to (step 310). (Step 308) The minimum prediction error value is updated to the calculated prediction error value. Next, go to (step 309). (Step 309) The motion vector corresponding to the minimum evaluation error value is updated. Next, go to (step 310). (Step 310) H_INTER is set to the horizontal motion vector.
Add VAL. Next, go to (step 311). (Step 311) It is checked whether or not the horizontal motion vector is larger than H_END. If so, go to (312). If not, go to (step 304). (Step 312) The horizontal motion vector is set to H_SRT. Add V_INTERVAL to the vertical motion vector. Next, go to (step 313). (Step 313) It is checked whether the vertical motion vector is larger than V_END. If it is larger, go to (314). If not, go to (step 304). (Step 314) The motion vector of the minimum evaluation error function is detected as a motion vector. The process ends.

The initial setting of the motion vector in FIG. 3 is 3
This is performed when the detection result is used when the motion vector value of the previous step is calculated in advance, such as the second step or the third step of the step method.

Next, a method of calculating a candidate vector by the search information calculator 102 will be described.

FIG. 4 is a diagram for explaining a method of calculating a candidate vector. Encoding is sequentially performed from the upper left block in the right direction, and when the processing of one horizontal macroblock line is completed, the processing of the next horizontal macroblock line is performed. Here, the average value of the motion vector for each horizontal value and vertical value of the upper block, the left block, the upper left block, and the lower right block, which are the four adjacent blocks of the coded block, is used as the prediction value. At this time, the capacity required for the motion vector memory 702 is the capacity corresponding to the motion vector of the number of blocks in the horizontal direction + 1.

As a method of calculating the candidate vector, the median value or the most frequent value of the motion vector may be used instead of the average value of the motion vector.

FIG. 5 shows the number of motion vector searches in the case of a method of calculating a candidate vector and performing two different searches (two-precision method using the calculation of a candidate vector). As shown in the figure, when the high-precision arithmetic unit 103 searches a candidate vector and eight pixels adjacent thereto, the number of searches is 9, and the low-precision arithmetic unit 104 has 9 searches, and the total number of searches is 18.

Here, the motion vector value at the search point of the high-precision arithmetic unit 103 becomes the motion vector value of the coded block when the block area is in a stationary state, pan / zoom, or moving object of the same speed. Is. The motion vector value at the search point of the low-precision arithmetic unit 104 becomes the motion vector value of the coding block in the case of a boundary between regions having different speeds or a region that does not exist in the previous frame. By providing the low-precision arithmetic unit 104, even when there is no motion vector in the area adjacent to the candidate vector, the motion vector is searched for in the entire search area, so that a large motion vector error is eliminated and deterioration of image quality can be prevented.

As described above, by using the video coding apparatus of the first embodiment, the processing of the high-precision arithmetic unit 103 and the low-precision arithmetic unit 104 is performed in parallel, so that the motion vector can be detected in one step. 3 because it becomes possible to do
The processing time can be reduced to about ⅓ as compared with the step method, or the increase of circuits by pipeline processing for avoiding it can be eliminated. Further, the number of motion vector searches can be reduced to 1/10 or less as compared with the full search method. Furthermore, by improving the accuracy of the candidate vector by the search information calculator 102 and by providing the low-precision calculator 104, it is possible to eliminate a large motion vector error and improve the image quality close to that of the full search method. .

Although the search accuracy is set to 1 pixel here, when it is set to 0.5 pixel accuracy, in addition to the 9 search points for high-precision motion vector detection in FIG. A motion vector is detected by performing a total of 17 search points including pixel search points in one step. That is, it is possible to detect the motion vector in one step even with an accuracy of 0.5 pixel.

(Embodiment 2) The reduction according to the first embodiment may impede the realization of the apparatus and may require further reduction.

Therefore, in the second embodiment, the motion vector can be detected by reducing the motion vector memory to a capacity necessary for the immediately preceding motion vector value and using the immediately preceding motion vector value as a candidate vector. A motion vector coding device will be described.

A video coding apparatus according to the second embodiment of the present invention will be described below with reference to the drawings.

FIG. 6 is a block diagram of the video coding apparatus in the second embodiment of the present invention. Here, the motion vector detector 10 of the video encoding device in the conventional example shown in FIG. 15 is used.
4 are different, their configurations are described.

In FIG. 6, the search area on the reference frame set in advance for detecting the motion vector is the first search area. A motion vector memory 601 stores the motion vector value of the immediately preceding coded block. 602 is a high-precision arithmetic unit, which is a motion vector memory 60.
A second search area including a block predicted from the motion vector value stored in 1 and smaller than the search area on the reference frame is set, and a candidate vector is set within the second search area with an accuracy of 1 pixel or less. A prediction error value is calculated for each candidate vector. A low-precision calculator 603 sets a candidate vector in the first search area on the reference frame with an accuracy of 2 pixels or more, and calculates a prediction error value for each candidate vector. Numeral 604 is a prediction error value comparator, which is a high-precision arithmetic unit 6
Prediction error value calculated by 02 and low-precision calculator 603
The motion vector value of the coding block is determined by comparing with the prediction error value calculated by.

FIG. 7 is a flow chart for explaining the operation of the detailed block shown in FIG. Below, the processing procedure is shown for the video data of the 1st frame after the 2nd frame. (Step 701) The number of pixels in the frame is 800 horizontal pixels / vertical 800 pixels, the block size is 16 horizontal pixels / 16 vertical pixels, the total number of horizontal blocks is 50, and the total number of vertical blocks is 50. The initial value is 0 and the vertical block initial value is 0. Next, go to (step 702). (Step 702) The horizontal direction of the motion vector memory 601 is set to HP, the vertical direction is set to VP, and both are initialized to 0. Next, (step 703) and (step 70
Go to 4). (Step 703) The high-precision arithmetic unit 602 detects a motion vector using the data of the encoded block and the data of the search area by the method shown in the flowchart of FIG.
When searching for 8 adjacent pixels centered on the value stored in the motion vector memory 601, the search range of the horizontal motion vector is HP-1 to HP + 1 and the search range of the vertical motion vector is VP-1 to VP-1. + VP + 1, the horizontal search accuracy is 1 pixel, and the vertical search accuracy is 1 pixel. Further, the minimum prediction error value and the initial setting value of its motion vector are not given. Here, the detected motion vector is HH in the horizontal direction and VH in the vertical direction, and the minimum prediction error value at that time is EH. Next, go to (step 705). (Step 704) The low-precision arithmetic unit 603 detects a motion vector using the data of the encoded block and the data of the search area by the method shown in the flowchart of FIG.
At this time, the search range of the horizontal motion vector is -4 to
+4, vertical motion vector search range is -4 to +
4. The search accuracy in the horizontal direction is 4 pixels, and the search accuracy in the vertical direction is 4 pixels. Also, the minimum prediction error value and the initial setting value of its motion vector are not given. Here, the detected motion vector is set to the horizontal direction HL and the vertical direction VL, and the prediction error value at that time is set to EL. Next, go to (step 705). (Step 705) Using the prediction error value comparator 604,
The prediction error value EH of the motion vector detected by the high precision calculator 602 is compared with the prediction error value EL of the motion vector detected by the low precision calculator 603. If EL is smaller than EH, HL and VL are set. If not, HH and VH are determined as motion vector values of the coding block. Next, go to (step 706). (Step 706) The motion vector determined by the prediction error value comparator 604 is stored in the motion vector memory 601.
Next, go to (Step 707). (Step 707) 1 in the horizontal block of the encoded frame
Is added. Next, go to (step 708). (Step 708) It is checked whether the horizontal block of the encoded frame is equal to 50, which is the total number of horizontal blocks. If they are equal, go to (step 709). If not, (step 703) and (step 704)
Go to (Step 709) The horizontal block of the encoded frame is set to 0.
Reset to. Add 1 to the vertical block of the encoded block. Next, go to (step 710). (Step 710) It is checked whether the vertical block of the encoded frame is equal to 50, which is the total number of vertical blocks. If they are equal, the process ends. If not, go to (step 702).

It should be noted that here, the high-precision arithmetic unit 602 and the low-precision arithmetic unit 603 respectively detect the motion vectors, and then the prediction error value comparator 604 compares the prediction error values with each other, thereby calculating the motion vector of the encoded block. Although the decision is made, the high-precision arithmetic unit 602 extracts all prediction error values of the coding block and the block corresponding to each search point, and the low-precision arithmetic unit 603 corresponds to the coding block and each search point. The prediction vector may be determined by extracting all the prediction error values with respect to the block to be encoded and comparing all the prediction error values with the prediction error value comparator 604.

As described above, by using the motion vector coding apparatus of the second embodiment, the immediately preceding motion vector value is used as the candidate vector, eliminating the candidate vector calculator and reducing the capacity of the motion vector memory to the immediately preceding position. It is possible to reduce only the motion vector values of the above, and the motion vector can be detected with a simpler configuration than the first embodiment. Furthermore, it is possible to reduce the number of constituent channels by thinning out 1/2 pixels in the vertical direction.

Pixel thinning can be used together as a further circuit reduction method. In particular, the vertical lines are decimated pixel by pixel (called vertical pixel ½ pixel decimation), and by detecting the motion vector using only one field of data, the pixel data in the search area required for motion vector detection is halved. Since the time required for memory access can be halved, in a circuit with a fast system clock, the number of constituent channels can be reduced as compared with the case where plural pixels are used when pixel thinning is not used.

In the video coding apparatus of this embodiment, the motion vector can be detected in one step by performing the processes of the high-precision arithmetic unit and the low-precision arithmetic unit in parallel. The processing time can be reduced as compared with, or the increase of circuits by pipeline processing for avoiding it can be unnecessary. In addition, the number of motion vector searches can be significantly reduced compared to the full search. Moreover, by using the previous motion vector value as a candidate vector,
It is possible to eliminate the candidate calculator and reduce the motion vector memory capacity to only the immediately preceding motion vector value. Furthermore, if the vertical ½ decimation is also used, it is possible to reduce the number of constituent channels when the plural channels are realized. Further, by using the high precision search and the low precision search together, it is possible to improve the image quality close to that of the full search method.

(Embodiment 3) In the conventional method, since the motion vector having the minimum prediction error value is determined as the motion vector of the coding block, both the coding block and the search area have the same brightness background wall. In the areas of the same pattern such as, the difference in the correlation value becomes small, and the motion vector value may be dispersed into various values.

In the methods shown in the first and second embodiments, the area to be searched in detail is determined by using the detected motion vector value of the block adjacent to the coded block. However, in such a method, in detecting motion vectors in the same pattern area, since the motion vector values of adjacent blocks are dispersed into various values, the motion vector of the coded block does not enter the area to be searched in detail. Since the probability increases, the prediction accuracy of the motion vector decreases at the boundary of the same pattern area, which causes a problem that the coding efficiency decreases.

As a motion vector coding method,
Coding efficiency is improved by using the method of variable length coding the difference value from the motion vector of the immediately preceding block. This is because the difference value is concentrated in the vicinity of 0 in the area other than the edge of the moving object in an image that often appears in a stationary state or during panning / zooming. Therefore, the difference value is variable length coded to reduce the code amount. This is because it is possible to

When such an encoding method is used, in the detection of the motion vector, in the case of the area of the same pattern,
There is almost no difference in the prediction error value for determining the prediction block, and the probability that the motion vector is detected at various values increases, so when the motion vector is differentially encoded,
There is a problem that the coding efficiency is reduced.

In the present embodiment, it is possible to increase the frequency of selecting the immediately preceding encoded motion vector value or a value in the vicinity thereof in the same pattern area.

A video coding apparatus according to the third embodiment of the present invention will be described below with reference to the drawings. FIG. 8 shows the third aspect of the present invention.
It is a block diagram of the video coding apparatus in an Example. Here, the motion vector memory 104 and the search information calculator 105 of the video encoding device in the first embodiment shown in FIG. 1 are used.
Since the function of the motion vector encoder 107 and the motion vector detector 106 are different, only those configurations are described.

In FIG. 8, the input video data divided into blocks is the coding block, and the area on the reference frame that performs block matching with the coding block is the search area. Further, when both the coded block and the search area include areas of the same brightness such as a background wall, the coded block is assumed to be the same pattern block.

In the figure, the search area on the reference frame set in advance for detecting the motion vector is defined as the first search area. A motion vector memory 801 stores the motion vector value of the immediately preceding coding block. 802 is a high-precision arithmetic unit, which is a motion vector memory 80.
A second search area that includes a block predicted from the motion vector value stored in 1 and is smaller than the search area on the reference frame is set, and a candidate vector is set in the second search area with one-pixel accuracy. A prediction error value is calculated for each candidate vector. A low-precision calculator 803 sets a candidate vector in the first search area on the reference frame with 4-pixel accuracy and calculates a prediction error value for each candidate vector. 80
A prediction error value comparator 4 compares the prediction error value calculated by the high-precision calculator 802 with the prediction error value calculated by the low-precision calculator 803 to obtain a motion vector value with the smallest prediction error. . Reference numeral 805 denotes a block classifier that classifies the coded blocks into the same pattern blocks or not by using the statistics of the prediction error value group. 806
Is a switch, and when the coded blocks are classified into the same pattern blocks by the block classifier 805, the motion vector value stored in the delay unit 801 is obtained. Otherwise, the minimum prediction obtained by the prediction error value comparator 804. The motion vector value corresponding to the error value is selected as the motion vector value of the coding block. Reference numeral 807 is a motion vector encoder, which performs variable length coding on the difference value of the motion vector values selected by the switch 806.

The operation of the video coding apparatus of the third embodiment constructed as described above will be described below. Here, only the motion vector detection and coding method, which is the main object of the present invention, will be described in detail.

FIG. 9 is a flow chart of the motion vector detecting and coding method of the video coding apparatus of the third embodiment. This illustrates the operation of the components in FIG. The processing procedure for the video data of the second frame is shown below. (Step 901) The number of pixels in a frame is 800 horizontal pixels / vertical 800 pixels, the block size is 16 horizontal pixels / 16 vertical pixels, the total number of horizontal blocks is 50, and the total number of vertical blocks is 50. The initial value is 0 and the vertical block initial value is 0. Next, go to (step 902). (Step 902) The motion vector memory 901 is initialized to 0 in both the horizontal and vertical directions. Next (Step 9
03) and (step 904). (Step 903) The high-precision arithmetic unit 802 sets a second search area from the value stored in the motion vector memory 801, sets a candidate vector in the second search area with one-pixel accuracy, and sets the candidate vector. A prediction error value is calculated for each. Next, go to (step 905). (Step 904) The low-precision calculator 803 sets a candidate vector in the first search area with 4-pixel accuracy and calculates a prediction error value for each candidate vector. Next, go to (step 905). (Step 905) Using the prediction error value comparator 804,
The motion vector having the minimum prediction error value is obtained from the prediction error value group obtained by the high precision arithmetic unit 802 and the low precision arithmetic unit 803. Next, go to (step 906). (Step 906) Using the block classifier 805, it is determined whether the blocks are the same pattern block. As the determination method, here, if the maximum value of the 21 prediction error values obtained by the low-precision arithmetic unit 803 is less than a certain threshold value, the same pattern is used.
Block lock. Next, go to (step 907). (Step 907) If the coded block is the same pattern block by using the switch 806, go to (step 908). Otherwise, (step 90
Go to 9). (Step 908) The value of the motion vector memory 801 is determined as the motion vector value of the coding block. Next, go to (step 910). (Step 909) The motion vector value of the position where the prediction error value obtained by the prediction error value comparator 804 is the minimum is determined as the motion vector value of the coding block. Next (Step 9
Go to 10). (Step 910) The motion vector value of the encoded block is stored in the motion vector memory 801. Next, go to (step 911). (Step 911) The motion vector encoder 807 is used to perform variable length coding on the difference value of the motion vector values selected by the switch 806. Next, go to (step 912). (Step 912) 1 in horizontal block of encoded block
Is added. Next, go to (step 913). (Step 913) It is checked whether the horizontal block of the encoded block is equal to 50, which is the total number of horizontal blocks. If they are equal, go to (step 914). If not, go to (step 903). (Step 914) Set the horizontal block of the encoded block to 0
Set to. Add 1 to the vertical block of the encoded block. Next, go to (step 915). (Step 915) It is checked whether the vertical block of the encoded block is equal to 50, which is the total number of vertical blocks. If they are equal, the process ends. If not,
Go to (step 902).

Here, the determination of the same pattern block will be described with reference to FIG. FIG. 14 is an explanatory diagram of the distribution of the prediction error value with respect to the motion vector value for determining the same pattern block using the prediction error value.

FIG. 14A shows motion vector detection in the case where the coding block and the search area are both in an area different from the area of the same pattern appearing on a wall of the same brightness in the background (hereinafter referred to as a normal area). Is shown. Also in FIG.
(B) represents the distribution of the prediction error value with respect to the motion vector value in the normal area. The motion vector value can be expressed in two dimensions, horizontal and vertical, but here the horizontal or vertical prediction error value including the minimum prediction error value is shown. In the normal area, the number of blocks that match the encoded block is one in the search area, so the prediction error value is small at the matching position, and the prediction error value is large at other locations.

On the other hand, FIG. 14C shows a motion vector in the case where the coding block and the search area are both in the same pattern area (hereinafter referred to as the same pattern area) appearing on a wall having the same brightness in the background. Indicates detection. Also in FIG.
(D) represents the distribution of prediction error values for motion vector values in the same pattern area. Since the coding block and the prediction block in the search area have the same brightness in the same pattern area, the prediction error value is a small value with respect to all motion vector values and its variation is small.

Therefore, in the same pattern area, since the motion vector is determined by a slight difference in prediction error value, the motion vector value is dispersed into various values, and the motion vector of the coding block is searched in detail. Since the probability of not falling into the range increases, the prediction accuracy of the motion vector decreases at the boundary of the same pattern area, and the coding efficiency decreases.
In addition, when the motion vector value is differentially encoded, the encoding efficiency decreases. Therefore, when it is determined that they are the same pattern block, the motion vector coding efficiency is improved by using the motion vector value of the immediately preceding coding block as the motion vector value.

In the block matching pattern shown in FIG. 10C, the following two features are found from the distribution of prediction error values shown in FIG. 10D. (A) The prediction error values for blocks in the search area are all small values. (B) The fluctuation of the prediction error value for the block in the search area is small.

Due to the feature (A), the following determination method can be used. (A) When the maximum prediction error value is equal to or less than the threshold value. (B) When the average value of the prediction error values is less than or equal to the threshold value.

Due to the feature (B), the following determination method can be used. (C) When the variance of the prediction error value is less than or equal to the threshold value. (D) The sum of absolute differences between the average of the prediction error values and each prediction error value is less than or equal to the threshold value. (E) When the difference between the maximum and minimum prediction error values is less than or equal to the threshold value.

It is possible to determine the same pattern block using these characteristics. Although (a) which can be realized with a simple configuration is used in this embodiment, other methods may be used.

As described above, by using the video coding apparatus of the third embodiment, the motion vector value of the coding block in the same pattern block can be made the motion vector value of the immediately preceding coding block. Since the probability that a motion vector is detected in a detailed search at the boundary of the same pattern area is high, the coding efficiency is improved. Further, the frequency of the motion vector difference value becoming 0 increases, and the motion vector difference value is variable-length coded, so that the motion vector coding efficiency can be improved.

Although the prediction error value used in the block classifier is obtained from the prediction error value group obtained from the low-precision arithmetic unit here, it may be used in the full search method or the first step of the three-step method. Good.

(Embodiment 4) In the third embodiment, all search areas are in the same pattern area, but in the coding blocks near the boundary of the same pattern area, the search areas are all in the same pattern area. Since it is not necessarily included, the same pattern block may not be determined by the method shown in the third embodiment. Therefore, in the fourth embodiment, even when the search area includes the boundary of the same pattern area, the same pattern is used.
A video encoding device capable of performing block determination will be described.

A video coding apparatus according to the fourth embodiment of the present invention will be described below with reference to the drawings. FIG. 11 is a block diagram of a video encoding apparatus according to the third embodiment of the present invention. Here, the motion vector memory 104 and the search information calculator 10 of the video encoding device in the first embodiment shown in FIG. 1 are used.
5 and the functions of the motion vector encoder 107 and the motion vector detector 106 are different, only their configurations are described.

In FIG. 11, input video data divided into blocks is a coding block, and a region on a reference frame for performing block matching with the coding block is a search region. Further, when both the coded block and the search area include areas of the same brightness such as a background wall, the coded block is assumed to be the same pattern block.

In the figure, the search area on the preset reference frame for detecting the motion vector is defined as the first search area. A motion vector memory 1101 stores the motion vector value of the immediately preceding coded block. 1102 is a high-precision arithmetic unit, which is a motion vector memory 1
A second search area including a block predicted from the motion vector value stored in 101 and smaller than the search area on the reference frame is set, and a candidate vector is set in the second search area with one-pixel accuracy, A prediction error value is calculated for each candidate vector. A low-precision calculator 1103 sets a candidate vector in the first search area on the reference frame with 4-pixel accuracy and calculates a prediction error value for each candidate vector. Reference numeral 1104 denotes a prediction error value comparator, which is a high-precision arithmetic unit 11
Prediction error value calculated by 02 and the low-precision calculator 110
The motion vector value with the smallest prediction error is obtained by comparing with the prediction error value calculated in 3. A block classifier 1105 classifies the coded blocks into the same pattern blocks or not by using the statistics of the prediction error value group. Reference numeral 1106 denotes a weighting unit, which is a block classifier 11
When the coded blocks are classified into the same pattern blocks by 05, the prediction error values obtained by the high-precision arithmetic unit and the low-precision arithmetic unit are weighted. A motion vector encoder 1107 performs variable length coding on the difference value between the motion vector values.

The operation of the video coding apparatus of the fourth embodiment constructed as described above will be described below. Here, only the motion vector detection and coding method, which is the main object of the present invention, will be described in detail.

FIG. 12 is a flowchart of the motion vector detection and coding method of the video coding apparatus according to the fourth embodiment. This illustrates the operation of the components in Figure 115. The processing procedure for the video data of the second frame is shown below. (Step 1201) Set the number of pixels in the frame to 800 horizontal pixels and 800 vertical pixels, and set the block size to 16 horizontal pixels.
Pixels and 16 pixels vertically, the total number of horizontal blocks is 50,
The total number of vertical blocks is 50, the initial value of horizontal blocks is 0, and the initial value of vertical blocks is 0. Next, go to (step 1202). (Step 1202) The motion vector memory 1201 is initialized to 0 in both the horizontal and vertical directions. Next, go to (step 1203) and (step 1204). (Step 1203) The high-precision arithmetic unit 1102 sets a second search area from the values stored in the motion vector memory 1101, sets a candidate vector in the second search area with one-pixel accuracy, and then sets the candidate vector. A prediction error value is calculated for each. Next, go to (step 1205). (Step 1204) The low-precision calculator 1103 sets a candidate vector in the first search area with 4-pixel accuracy, and calculates a prediction error value for each candidate vector. Next, go to (step 1205). (Step 1205) The block classifier 1105 is used to determine whether the blocks are the same pattern block. As the determination method, here, 21 obtained by the low-precision arithmetic unit 1103 is used.
Of the prediction error values, the maximum of the prediction error values with the same horizontal component or vertical component is below a certain threshold,
Use the same pattern block. Next (step 120
Go to 6). (Step 1206) If the coded blocks are the same pattern block, go to (Step 1207). If not, go to (step 1208). (Step 1207) The weighting device 1106 is used to weight the prediction error value obtained by the high-precision arithmetic unit so as to be small. Next, go to (step 1208). (Step 1208) The motion vector value of the position where the prediction error value obtained by the prediction error value comparator 1104 is the minimum is determined as the motion vector value of the coding block. Next, go to (step 1209). (Step 1209) The motion vector value of the encoded block is stored in the motion vector memory 1101. Next, go to (step 1210). (Step 1210) The motion vector encoder 1107 is used to perform variable length coding on the difference value between the motion vector values.
Next, go to (step 1211). (Step 1211) 1 is added to the horizontal block of the encoded block. Next, go to (step 1212). (Step 1212) It is checked whether the horizontal block of the encoded block is equal to 50, which is the total number of horizontal blocks. If they are equal, go to (step 1213). If not, go to (Step 1203). (Step 1213) The horizontal block of the encoded block is set to 0. Add 1 to the vertical block of the encoded block. Next, go to (step 1214). (Step 1214) It is checked whether the vertical block of the encoded block is equal to 50, which is the total number of vertical blocks. If they are equal, the process ends. If not,
Go to (step 1202).

Here, the determination of the same pattern block will be described with reference to FIGS. 13 and 14. Figure 13
The coded block is a block in the same pattern area,
It is an explanatory view of distribution of the prediction error value with respect to the motion vector value when the search area extends over the same pattern area and the normal area. When the coding block is a block in the same pattern area and the search area extends over two different pattern areas, the prediction error value distribution shown in FIG. Since it is possible to determine the same pattern block, it is not mentioned here.

On the other hand, FIG. 14 is an explanatory view of the distribution of the prediction error value with respect to the motion vector value when the coding block includes two different identical pattern areas. If the coding block includes the same pattern area and the normal area,
It is possible to determine the prediction block from the difference in the partial areas of the normal area.

FIG. 13A shows motion vector detection when the coded block is a block in the same pattern area and the search area extends over the same pattern area and the normal area. FIG. 13B shows the distribution of prediction error values for horizontal motion vector values. The motion vector value can be expressed in two dimensions, horizontal and vertical, but here the prediction error value including the minimum motion vector value is shown. When the prediction block is in the same pattern area, the size and fluctuation of the prediction error value are small. In addition, when the prediction block is in the normal area, the size and variation of the prediction error value are large.

FIG. 13C shows the distribution of the prediction error value with respect to the motion vector value in the vertical direction when the prediction block is in the same pattern area. In this case, since the brightness of the coding block is the same as that of the prediction block, the size and fluctuation of the prediction error value are small.

FIG. 13D shows the distribution of the prediction error value with respect to the vertical motion vector value when the prediction block is in the normal area. In this case, since the correlation between the coding block and the prediction block is low, the magnitude of the prediction error value and its variation are large. FIG. 13E shows the distribution of the prediction error value in the vertical direction when the prediction block extends over the same pattern area and the normal area. Since the correlation is high in the same pattern area in the prediction block and the correlation is low in the normal area, the distribution of the prediction error value is a value between FIG. 13 (c) and FIG. 13 (d).

On the other hand, FIG. 14A shows the motion vector detection in the case where both the coding block and the search area extend over two different identical pattern areas (called the same pattern area 1 and the same pattern area 2). Shows. FIG. 14 (b)
Represents the distribution of the prediction error value with respect to the horizontal motion vector value including the minimum prediction error value. First, when the prediction block is in the same pattern area 1, the luminance difference is small in the partial area that is the same pattern area 1 in the coding block. On the other hand, in the partial areas that are the same pattern area 2 in the coded block, the difference in brightness is large, but since they are the same pattern area, the difference in brightness is almost constant for each prediction block. Therefore, the prediction error value is almost constant.

Next, the prediction block has the same pattern area 2
In the case of, the same pattern area 2 in the encoded block
The luminance difference is small in the partial area. On the other hand, the luminance difference is large in the partial area that is the same pattern area 1 in the encoded block, and is almost constant for each prediction block.
Therefore, the prediction error value is almost constant. Further, when the prediction block includes two identical pattern areas, the motion vector value increases or decreases with the motion vector value at the position where the prediction error value is the smallest in the partial area of the same pattern area different from the coding block. Increases in proportion to. Therefore, the prediction error value increases in proportion to the motion vector value.

FIG. 14C shows the detection of the motion vector in the vertical direction when the prediction areas are in the same pattern area 1. In this case, since the difference value between the partial area of the same pattern area 2 and the same pattern area 1 of the encoded block is substantially constant at a certain size, the size of the prediction error value is also constant. FIG. 14D shows the distribution of the prediction error value with respect to the vertical motion vector value when the prediction region is in the same pattern region 2. Also in this case, FIG.
Similar to (c), the magnitude of the prediction error value is almost constant. FIG. 14E shows the distribution of vertical prediction error values including the minimum prediction error value. Since the difference in brightness between all the prediction blocks and the coding blocks is small, the size and fluctuation of the prediction error value are small.

Therefore, when the boundary of the same pattern area is horizontal or vertical, the same pattern block is determined for a plurality of prediction error value groups having the same horizontal or vertical component of the motion vector, and at least one or more of them are determined. If the group of prediction error values satisfies the judgment of the same pattern block, it is possible that the coding block includes the same pattern area. Further, this condition is satisfied even when the search area shown in FIG. 15 is included in the same pattern area.

The following methods can be considered as weighting methods. (1) Weighting is performed so that only the immediately preceding motion vector value is used. (2) Weighting is performed so as to reduce the prediction error value of the immediately preceding motion vector value. (3) Weighting is performed so that only the prediction error value obtained by the high-precision arithmetic unit 1102 is used. (4) Weighting is performed so as to reduce the prediction error value obtained by the high precision arithmetic unit 1102.

In (1), the differential coding efficiency is improved, but when the prediction error value of the position corresponding to the immediately preceding motion vector value is large, the coding efficiency is deteriorated.

In (2), by weighting the prediction error value of the position corresponding to the immediately preceding motion vector, the probability that the immediately preceding motion vector value is selected is increased, and if the prediction error value is large, it is not selected. Is the way to. In this case, a multiplication function for weighting is necessary.

In (3), the method of using only the prediction error value adjacent to the position corresponding to the immediately previous motion vector value obtained by the high-precision arithmetic unit 1102 improves the differential coding efficiency and improves the prediction error. By using the value, the coding efficiency can be improved.

In (4), weighting is performed so that the coding efficiency can be improved more than in (1). However, in this case, a multiplication function for weighting is required.

As described above, by using the video coding apparatus of the fourth embodiment, it is possible to determine the motion vector value of the coding block immediately before by judging the same pattern block in the coding block including the same pattern area. It becomes possible to set the motion vector value of the coding block of, and the frequency of the difference value of the motion vector becoming 0 increases. Therefore, by encoding the difference value of the motion vector by variable length coding, the coding efficiency of the motion vector can be improved.

Further, as compared with the third embodiment, it is possible to determine the same pattern block even for the coded blocks in the vicinity of the boundary of the same pattern area, and the detection accuracy of the same pattern block can be improved. Since it increases, the coding efficiency of the motion vector is further improved.

Although it is assumed here that the boundary between the same pattern area and the normal area is horizontal or vertical, various kinds of processing can be performed by using a prediction error value group on a diagonal line or an arbitrary curve. It is possible to accommodate boundaries.

[0088]

The video coding apparatus of the present invention can detect a motion vector in one step by performing the processing of the high-precision arithmetic unit and the processing of the low-precision arithmetic unit in parallel. Will processing time be reduced compared to the law?
It is not necessary to increase the number of circuits by pipeline processing to avoid this. In addition, the number of motion vector searches can be significantly reduced compared to the full search. Furthermore, by using the high-precision search and the low-precision search together, it is possible to improve the image quality close to the full search method.

[Brief description of drawings]

FIG. 1 is a block diagram of a video encoding device according to a first embodiment of the present invention.

FIG. 2 is a flowchart of the video encoding device according to the first embodiment of the present invention.

FIG. 3 is a flowchart of motion vector search according to the first embodiment of the present invention.

FIG. 4 is an explanatory diagram of a candidate vector extraction method according to the first embodiment of the present invention.

FIG. 5 is an explanatory diagram of motion vector search according to the first embodiment of the present invention.

FIG. 6 is a block diagram of a video encoding device according to a second embodiment of the present invention.

FIG. 7 is a flowchart of the video encoding device according to the second embodiment of the present invention.

FIG. 8 is a block diagram of a video encoding device according to a third embodiment of the present invention.

FIG. 9 is a flowchart of the video encoding device according to the third embodiment of the present invention.

FIG. 10 is a diagram showing distribution of prediction error values with respect to motion vector values in the third embodiment of the present invention.

FIG. 11 is a block diagram of a video encoding device according to a fourth embodiment of the present invention.

FIG. 12 is a flowchart of a motion compensation encoding device according to a fourth embodiment of the present invention.

FIG. 13 is a diagram showing a distribution of prediction error values with respect to motion vector values at a region boundary when coding blocks are in the same pattern region according to the fourth embodiment of the present invention.

FIG. 14 is a diagram showing a distribution of a correlation error value with respect to a motion vector value when a coding block is on the boundary of two identical pattern areas in the fourth embodiment of the present invention.

FIG. 15 is a block diagram of a video encoding device in a conventional example.

FIG. 16 is an explanatory diagram of a motion vector search area in a conventional example.

FIG. 17 is an explanatory diagram of a conventional full search method.

FIG. 18 is an explanatory diagram of a three-step method in a conventional example.

[Explanation of symbols]

 101 Motion Vector Memory 102 Search Information Calculator 103 High Precision Calculator 104 Low Precision Calculator 105 Prediction Error Value Comparator

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Takumi Hasebe 1006 Kadoma, Kadoma, Osaka Prefecture Matsushita Electric Industrial Co., Ltd. (72) Hiroshi Nishikawa, 1006 Kadoma, Kadoma City, Osaka Matsushita Electric Industrial Co., Ltd.

Claims (13)

[Claims] 1. A motion for storing a motion vector value of a coded block in which a motion vector is detected, when a search region on a reference frame set in advance for detecting a motion vector is used as a first search region. Vector memory,
A candidate vector calculator that calculates one candidate vector from a motion vector value stored in the motion vector memory according to a predetermined arithmetic expression, and a block that is predicted by the candidate vector and is smaller than a search area on a reference frame A high-precision calculator that sets a second search area, sets a candidate vector in the second search area with an accuracy of one pixel or less, and calculates a prediction error value for each candidate vector; A low-precision calculator that sets a candidate vector in the first search area with an accuracy of 2 pixels or more and calculates a prediction error value for each candidate vector; and a prediction error value calculated by the high-precision calculator. A prediction error value comparator for determining a motion vector value of a coding block by comparing with a prediction error value calculated by the low precision arithmetic unit Video encoding apparatus for. 2. When a search area on a reference frame set in advance for detecting a motion vector is used as a first search area, a reference including a block predicted by a motion vector value of an immediately preceding coding block is included. A high-precision calculation that sets a second search area smaller than the search area on the frame, sets a candidate vector in the second search area with an accuracy of one pixel or less, and calculates a prediction error value for each candidate vector. And a low-precision calculator that sets a candidate vector in the first search area on a reference frame with an accuracy of 2 pixels or more and calculates a prediction error value for each candidate vector, and the high-precision calculator. A prediction error value comparator for determining the motion vector value of the coding block by comparing the calculated prediction error value with the prediction error value calculated by the low-precision calculator is provided. Video encoding apparatus according to claim and. 3. The high-precision searcher and the low-precision searcher calculate a prediction error value using a coded block and a search region in which vertical pixels are thinned out pixel by pixel. Alternatively, the video encoding device according to claim 2. 4. When the input video data divided into blocks is used as a coding block and a region on a reference frame for performing block matching with the coding block is used as a search region, the coding block and the block of the search region Prediction error value comparator for obtaining the minimum value of the prediction error value group with the group,
A block classifier that classifies the coded blocks into patterns 1 and 2 using the statistics of the prediction error value group and a block classifier that is adjacent when the coded blocks are classified into the pattern 1 When the motion vector value of the block is classified into the pattern 2, a switch for selecting the motion vector value corresponding to the minimum prediction error value obtained by the prediction error value comparator as the motion vector value of the coding block is selected. A video encoding device provided with. 5. When the input video data divided into blocks is a coding block, a region on a reference frame for performing block matching with the coding block is a search region, and the search region is a first search region. , Setting a second search area smaller than the search area on the reference frame including the block predicted by the motion vector value of the adjacent block, and setting the candidate vector in the second search area with an accuracy of 1 pixel or less And a high-precision calculator for calculating a prediction error value for each candidate vector, and a prediction error for each candidate vector set in the first search area on the reference frame with an accuracy of 2 pixels or more. Comparing a prediction error value calculated by the high precision calculation unit with a prediction error value calculated by the low precision calculation unit And more predictable error value comparator prediction error values determine the smallest block, the prediction pattern 1 the encoded block using statistics of error value group and pattern 2
And a motion vector value stored in the motion vector memory when the coded block is classified into the pattern 1 by the block classifier, and a motion vector value stored in the pattern 2 when the coded block is classified into the pattern 1. A video coding apparatus comprising a switch for selecting a motion vector value corresponding to a minimum prediction error value obtained by a prediction error value comparator as a motion vector value of the coding block. 6. When the input video data divided into blocks is used as a coding block and a region on a reference frame for performing block matching with the coding block is used as a search region, the coding block and the block of the search region A block classifier that classifies the coded blocks into pattern 1 and pattern 2 by using the statistics of the prediction error value group with the group;
A weighter for weighting the prediction error value group when the coded block is classified into the pattern 1 by the block classifier, and a prediction for obtaining the minimum value of the prediction error value group weighted by the weighter An image coding apparatus comprising an error value comparator. 7. When input video data divided into blocks is a coding block, a region on a reference frame for performing block matching with the coding block is a search region, and the search region is a first search region. , Setting a second search area smaller than the search area on the reference frame including the block predicted by the motion vector value of the adjacent block, and setting the candidate vector in the second search area with an accuracy of 1 pixel or less And a high-precision calculator for calculating a prediction error value for each candidate vector, and a prediction error for each candidate vector set in the first search area on the reference frame with an accuracy of 2 pixels or more. Patterning the encoded block using a low-precision arithmetic unit that calculates a value, and statistics of the prediction error value group of the encoded block and the block group of the search region 1 and a pattern 2, a block classifier, a weighter that weights the prediction error value group when the block is classified into the pattern 1, and a weighter that weights the prediction error value group. And a prediction error value comparator for obtaining the minimum value of the prediction error value group. 8. The video according to claim 6, wherein the weighter reduces the weighting of the prediction error value at the position corresponding to the candidate vector set by the motion vector value of the adjacent block. Encoding device. 9. The video coding apparatus according to claim 6, wherein the weighting unit reduces the weighting of the prediction error value calculated by the high precision search unit. 10. A motion vector encoder for variable length coding a difference value between motion vector values, according to claim 4, claim 5, claim 6, or claim 7. Video encoding device. 11. A partial prediction error is defined when one partial area or a plurality of overlapping partial areas are set in a search area and the prediction error value group in the partial area is set as a partial prediction error value group. The classification is performed on a value group, and the coded block is patterned when the partial prediction error value group of at least one or more partial regions is classified into the pattern 1.
8. The coding block classifier for classifying the coding block into the class 1 is provided, claim 4, claim 5, claim 6, and claim 7.
The video encoding device according to any one of 1. 12. A partial area or a plurality of partial areas that can overlap each other are formed in an area search area, which is composed of a block group in which the horizontal value or the vertical value of a motion vector corresponding to a prediction error value is the same value. When setting the prediction error value group in the partial region as a partial prediction error value group, the classification is performed on the partial prediction error value group, and the partial prediction of at least one or more partial regions is performed. 6. The coding block classifier for classifying the coding block into the pattern 1 when the error value group is classified into the pattern 1, wherein the coding block classifier is provided. The video encoding device according to claim 6 or 7. 13. When a prediction error value group or a partial prediction error value group for a coded block is a classification prediction error value group,
As a statistic, a statistic indicating the size of the classification prediction error value group including at least the maximum value or the average value of the classification prediction error value group, or at least the variance value or the average value of the classification prediction error value group. Using a statistic representing the variation of the classification prediction error value group including the difference between the maximum value and the minimum value of the classification prediction error value group or the sum of absolute values of the differences between the prediction error values, 5. The coding block classifier which sets the pattern 1 when the value is less than or equal to a set value and sets the pattern 2 when the value is less than the set value, claims 4, 5, 6 and 6. 7. The video encoding device according to any one of 7.