JPH10224796A - Motion vector detector - Google Patents

Abstract

PROBLEM TO BE SOLVED: To attain frame prediction, field prediction and dual prime prediction also in frame structure without increasing a circuit scale by switching the time slot of motion vector retrieval. SOLUTION: The motion vector detector is constituted of a motion vector rough retrieval part 1, a motion vector rough result 11, an input picture buffer 12, an encoding control part 2, a motion vector accurate retrieval part 4, a motion vector detection part 43, and so on. In the constitution, the rough retrieval part 1 retrieves the accuracy of two horizontal pixels and one vertical pixel. Then the accurate retrieval part 4 retrieves the accuracy of one horizontal pixel and one vertical pixel around the periphery of an obtained vector. Then horizontal/vertical 1/2 pixel accuracy is retrieved on the periphery of the vector obtained by the retrieval of the horizontal/vertical one pixel accuracy. Thus retrieval of three steps is executed. The data of each macro block to be a unit for finding out a motion vector are sent from the buffer 12 to an input picture buffer 46 in order to reduce memory access.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital image signal encoding apparatus for encoding a digital image signal by variable-length high-efficiency encoding, in which a motion vector necessary for reducing the redundancy in the time axis direction is obtained. The present invention relates to a motion vector detection device used for detection.

[0002]

2. Description of the Related Art Predictive coding of a moving image using motion compensation such as MPEG2, which is an international standard, can provide high compression efficiency, but is widely used to detect a motion vector used for motion compensation. It is necessary to perform pattern matching in the range. Therefore, in terms of memory access and circuit scale, the ratio occupied by motion vector detection is very large. Therefore, as described in Japanese Patent Application Laid-Open No. Hei 5-236466, a two- (multi) -step search in which a wide range is searched with low accuracy and then a narrow range is searched with high accuracy, or reference data for generating a prediction error is used. And the memory access is shared.

However, in the prior art, in order to encode a macroblock which is a unit of motion vector detection, an intra-picture coded image (hereinafter, I-picture) and a forward prediction coded image (hereinafter, P-picture) are used. In addition, a time slot is provided for each of at least three types of predicted image signals of a bidirectional predictive coded image (hereinafter referred to as a B-picture) to detect a motion vector. The prediction in the forward direction and the backward direction is performed in the time division, and the time division processing of the time slot is not performed in the I and P-pictures.

[0004] Therefore, in order to efficiently encode an interlaced image such as MPEG2, frame prediction and field prediction in a frame structure in which encoding is performed in frame units, and field prediction and field prediction in a field structure in which encoding is performed in field units. When it is necessary to detect a plurality of types of motion vectors, such as 16 × 8 MC, and further, a frame structure can derive four vectors and a field structure can derive two vectors. I cannot respond to prime prediction.

[0005]

SUMMARY OF THE INVENTION It is an object of the present invention to provide a method for performing frame prediction and field prediction in a frame structure, field prediction and 16 × 8 MC in a field structure, and dual prime (time) in a time slot allocated without waste. Dual-prime) prediction is also realized without increasing the circuit scale.

[0006]

In order to achieve the above object, according to the present invention, a reference image and an input block image used for prediction are inputted, and an optimum motion vector is detected within a predetermined search range. A search unit and a search for a motion vector using the search unit.
The first search in the bidirectional prediction coded image is performed by searching for a forward prediction motion vector, and the second search is performed by the search means so as to search for a backward prediction motion vector. In the forward prediction coded image, a motion vector of forward prediction is searched for the first time, a main vector of dual prime prediction is generated from the motion vector obtained in the first search, and a dual prime prediction is performed for the second time. And a search control means for controlling the search means so as to search for a small difference vector.

[0007]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a schematic block diagram of an embodiment of a motion vector detecting apparatus according to the present invention. The motion vector coarse search section 1, the motion vector coarse search result 11, the input image buffer 12, the encoding control section 2 , Macro block address 2
1, picture type 22, GOP (Group of
Pictures 23, memory 3, motion vector dense search unit 4, vector switching SW41, dual-prime (Dual-prime) determination unit 42, motion vector detection unit 43, input image switching SW44, interpolation image creation unit 4
5, input image buffer 46, reference image read address generation 47, reference image buffer 48, vector output terminal 49
Consists of

In the present embodiment, the coarse motion vector search unit 1 performs a horizontal two-pixel and one-pixel vertical search, and the motion vector fine search unit 4 searches for a horizontal and one-pixel accuracy around the obtained vector. A three-stage search is performed in which a search with a horizontal and vertical 1/2 pixel precision is performed around a vector obtained by a search with a horizontal and vertical 1 pixel precision. In order to reduce memory access, data for a macroblock, which is a unit for obtaining a motion vector, is sent from the input image buffer 12 to the input image buffer 46.

A case where the image to be encoded is a B-picture will be described.

From the motion vector coarse search determination result 11, forward prediction and backward (b)
ackward) The vector of the prediction is sequentially and time-division-divided from the forward prediction by the motion vector
Output to W41. Since the signal from the picture type 22 is a signal indicating a B-picture, the dual prime determination unit 42 does not depend on the signal from the GOP configuration 23.
Outputs a signal that cannot be dual primed.

The vector switching SW 41 outputs a forward prediction vector to the reference image read address generation 47 based on the signal from the dual prime determination section 42 and the count value of the number of fine searches from the motion vector detection section 43.
The reference image read address generator 47 receives this vector and outputs the reference image read address to the memory 3.

In the reference image for searching for a motion vector stored in the memory 3, data for a search range is read into a reference image buffer 48. The input image switching SW 44 outputs the output from the input image buffer 46 to the motion vector detection unit 43 based on the signal from the dual prime determination unit 42 and the fine search count value from the motion vector detection unit 43. The reference image buffer 48 sends data of the search range to the motion vector detection unit 43.

Here, the motion vector detecting section 43 performs matching between the two, obtains a vector with horizontal and vertical 1-pixel accuracy when the difference is the smallest, and outputs the vector to the reference image buffer 48 and the vector switch SW41. The vector switching SW 41 uses the signal from the dual prime determination unit 42 and the count value of the number of fine searches from the motion vector detection unit 43 to generate a backward prediction vector as a reference image read address generation 47.
Output to

The reference image read address generator 47 receives this vector and outputs the read address of the reference image to the memory 3. Among the reference images for searching for the motion vector stored in the memory 3, the data for the search range is read into an area of the reference image buffer 48 other than the area where the forward prediction reference image is written.

The reference image buffer 48 stores, within the reference image for searching for a motion vector for forward prediction, data corresponding to a search range necessary for searching with 1/2 pixel accuracy around the obtained vector. Send to 43. The input image switching SW 44 outputs the output from the input image buffer 46 to the motion vector detection unit 43 based on the signal from the dual prime determination unit 42 and the fine search count value from the motion vector detection unit 43. Here, the motion vector detection unit 43 performs matching between the two, obtains a vector with horizontal and vertical 1/2 pixel accuracy when the difference is the smallest, and outputs the vector from the vector output terminal 49.

A similar process is performed on the vector for backward prediction. Where the data for the search range of the backward prediction is read into the reference image buffer 48, the data of the forward prediction of the next macroblock is read.

FIG. 2 shows the above operation as time on the horizontal axis. FIG. 2 shows a processing time slot and a memory access time slot in a B-picture. In the figure, an integer coarse search represents the time during which a search with horizontal and vertical 1-pixel accuracy is being processed, and a half coarse search is horizontal and vertical 1/2.
Represents the time during which the search for pixel accuracy is being processed. The memory access indicates a period during which the memory access is being performed for reading the reference image, and the hatched portion in the figure indicates that the memory access is not performed.

Next, the case where the image to be encoded is a P-picture and m is other than 1 will be described.

In the case of a frame structure, from the motion vector coarse search determination result 11, vectors for frame prediction and field prediction are output to the motion vector detection unit 43 and the vector switching SW 41 in a time-division manner in order from the frame prediction. Since the signal from the picture type 22 is a signal indicating a P-picture, the signal from the GOP structure 23 is m = 1.
At times other than (where m is the period at which an I or P-picture appears in the GOP), the dual prime determination unit 42 outputs a signal indicating that dual prime is not possible.

The vector switching SW 41 outputs a frame prediction vector to the reference image read address generation 47 based on the signal from the dual prime determination unit 42 and the count value of the number of fine searches from the motion vector detection unit 43.

The reference image read address generator 47 receives this vector and outputs the read address of the reference image to the memory 3. In the reference image for searching for the motion vector stored in the memory 3, the data for the search range is read into the reference image buffer 48. Input image switch SW44
Is calculated from the signal from the dual prime determination unit 42 and the count value of the number of dense searches from the motion vector detection unit 43.
The output from the input image buffer 46 is output to the motion vector detection unit 43. The reference image buffer 48 sends data of the search range to the motion vector detection unit 43.

The motion vector detecting section 43 performs matching between the two, obtains a vector having a horizontal and vertical 1-pixel accuracy when the difference is the smallest, and outputs the vector to the reference image buffer 48 and the vector switch SW41. The vector switching SW 41 uses the signal from the dual prime determination unit 42 and the count value of the number of fine searches from the motion vector detection unit 43 to convert the field prediction vector into a reference image read address generation 47.
Output to

The reference image read address generator 47 receives this vector and outputs the read address of the reference image to the memory 3. In the reference image for searching for the motion vector stored in the memory 3, the data for the search range is read into an area of the reference image buffer 48 which is different from the area where the frame prediction reference image is written.

The reference image buffer 48 stores, within a reference image for searching a motion vector for frame prediction, data corresponding to a search range necessary for searching with 1/2 pixel accuracy around the obtained vector. Send to 43. The input image switching SW 44 outputs the output from the input image buffer 46 to the motion vector detection unit 43 based on the signal from the dual prime determination unit 42 and the fine search count value from the motion vector detection unit 43.

The motion vector detection unit 43 performs matching between the two, and determines the horizontal / vertical 1 / D when the difference is the smallest.
A vector with two-pixel accuracy is obtained and output from the vector output terminal 49.

Similar processing is performed on the field prediction vector. Where data for the search range of the field prediction is being read into the reference image buffer 48, the frame prediction data of the next macroblock is read.

FIG. 3 shows the above operation as time on the horizontal axis. FIG. 3 shows a processing time slot and a memory access time slot when the picture is a P-picture and m is other than 1. In the figure, the integer coarse search is horizontal and vertical 1
The half image coarse search indicates the time during which the search for the pixel accuracy is being processed, and the half image coarse search indicates the time during which the search with the horizontal and vertical half pixel accuracy is being processed. The memory access indicates a period during which the memory access is being performed for reading the reference image, and the hatched portion in the figure indicates that the memory access is not performed.

The next image to be encoded is a P-picture,
The case where m = 1 will be described.

In the case of the frame structure, the frame prediction and field prediction vectors are output to the motion vector detection unit 43 and the vector switching SW 41 in time division from the frame prediction in the motion vector coarse search determination result 11. Since the signal from the picture type 22 is a signal indicating a P-picture, the signal from the GOP structure 23 is m = 1.
In the case of (m is a period in which an I or P-picture appears in a GOP), the dual prime determination unit 42 outputs a signal indicating that dual prime is possible.

The vector switching SW 41 outputs a frame prediction vector to the reference image read address generation 47 based on the signal from the dual prime determination section 42 and the count value of the number of fine searches from the motion vector detection section 43. The reference image read address generator 47 receives this vector and outputs the reference image read address to the memory 3. In the reference image for searching for the motion vector stored in the memory 3, the data for the search range is read into the reference image buffer 48.

The input image switching SW 44 receives the signal from the dual prime judging section 42 and the count value of the number of fine searches from the motion vector detecting section 43 and outputs the input image buffer 4.
6 is output to the motion vector detection unit 43. The reference image buffer 48 sends data of the search range to the motion vector detection unit 43.

The motion vector detector 43 performs matching between the two, obtains a vector with horizontal and vertical 1-pixel accuracy when the difference is the smallest, and outputs the vector to the reference image buffer 48 and the vector switch SW41.

The vector switch SW 41 outputs a vector having a horizontal and vertical 1-pixel accuracy to the reference image read address generation 47 based on the signal from the dual prime determination section 42 and the count value of the number of fine searches from the motion vector detection section 43.

The reference image read address generator 47 uses this vector as a main vector used for the prediction of the same parity (same parity) in the dual prime prediction, and sets this vector to a different parity (opposition).
For the prediction of (parity), scaling is performed, and the read address of the reference image is output to the memory 3 based on the scaled vector.

In the reference image for searching for the motion vector stored in the memory 3, the data for the search range is read into an area of the reference image buffer 48 which is different from the area where the frame prediction reference image is written. . Reference image buffer 4
8 sends to the motion vector detection unit 43 data of a search range necessary for searching with 1/2 pixel accuracy around the obtained vector in the reference image for searching for a motion vector for frame prediction. .

The input image switching SW 44 receives the signal from the dual prime judging section 42 and the count value of the number of fine searches from the motion vector detecting section 43 and outputs the input image buffer 4.
6 is output to the motion vector detection unit 43.

The motion vector detecting section 43 performs matching between the two, and determines the horizontal / vertical 1 / D when the difference is the smallest.
A vector with two-pixel accuracy is obtained and output from the vector output terminal 49. The interpolation image creation unit 45 creates a new interpolation original image for dual prime prediction from the data of the same parity from the reference image buffer 45 and the data from the input image buffer 46, and sends it to the input image switching SW 44.

The input image switching SW 44 is provided with an interpolation image generation unit 45 based on the signal from the dual prime determination unit 42 and the count value of the number of fine searches from the motion vector detection unit 43.
Is output to the motion vector detection unit 43. The reference image buffer 48 sends data of the search range to the motion vector detection unit 43.

Here, the motion vector detecting section 43 takes matching between the two in a range of 1/2 pixel in the upper, lower, left and right directions, and performs a differential motion vector (Differential Motion Video) with horizontal and vertical 1/2 pixel precision when the difference is the smallest.
ctor, DMV) and outputs it from the vector output terminal 49.

Next, the frame prediction vector of the next macroblock is input to the vector switch SW41 from the motion vector coarse search determination result 11. The vector switching SW 41 outputs a frame prediction vector to the reference image read address generation 47 based on the signal from the dual prime determination unit 42 and the fine search count value from the motion vector detection unit 43. The reference image read address generator 47 receives this vector and outputs the reference image read address to the memory 3.

In the reference image for searching for the motion vector stored in the memory 3, data of a search range is written in the reference image buffer 48 as a reference image of a different parity (opposite parity) for dual prime prediction. Read into an area different from the area where

FIG. 4 shows the above operation as time on the horizontal axis. FIG. 4 shows a processing time slot and a memory access time slot when a picture is a P-picture and m = 1. In the figure, the integer coarse search represents the time during which the horizontal and vertical 1-pixel precision search is being processed, the half coarse coarse search represents the time during which the horizontal and vertical 1 / 2-pixel precision search is being processed, and a differential motion vector. The (DMV) search indicates the time during which the search in the dual prime prediction is being processed, and the shaded portion in the figure indicates that the search processing has not been performed.

The memory access indicates a period during which the memory access is being performed for reading the reference image, and the hatched portion in the figure indicates that the memory access is not being performed.

By switching the time slots in this manner, it is possible to realize the frame prediction and the field prediction in the frame structure, the field prediction and the 16 × 8 MC in the field structure, and also the dual prime prediction without increasing the circuit scale. Can be done.

In this embodiment, only the case of the frame structure has been described. However, the same can be applied to the field structure. When m = 1 in the P-picture, the field search is performed in the first search, and the first search is performed. Using the obtained vector as a main vector for dual prime prediction, a differential motion vector (DMV) search is performed in the second search.

Next, referring to FIG.
3 will be described in detail.

FIG. 5 is a schematic configuration diagram of an embodiment of a motion vector detecting section in a motion vector detecting device using the present invention. In FIG. 5, the same components as those in FIG. 1 are denoted by the same reference numerals. Numeral 431 is a search area limiting unit, 432 is a prediction error calculation unit, 433 is a coding mode control unit, 434 is a prediction error comparison unit, and 435 is a motion vector output terminal.

An input image switching SW is provided to the prediction error calculating section 432.
The data of a macro block which is a unit for obtaining a motion vector input from 44 and the data of a search range input from a reference image buffer 48 are input. The encoding mode control unit 433 includes the dual prime determination unit 42
And a signal indicating whether to perform a horizontal / vertical 1-pixel precision search, a horizontal / vertical 1 / 2-pixel precision search, or a dual prime search, based on the signal from the internal search and the fine search count value by the internal counter. Output to the calculation unit 432 and the prediction error comparison unit 434.

When the signal from the encoding mode control unit 433 indicates a search with one pixel in the horizontal and vertical directions, the prediction error calculation unit 432 moves the reference image up, down, left, and right within the search range, and The absolute value difference for each pixel is calculated, the absolute value difference sum per processing block is obtained, and the vector at that time and the absolute value difference sum are output to the prediction error comparison unit 434.

When the signal from the encoding mode control unit 433 indicates a search with horizontal and vertical half-pixel accuracy and a dual prime search, the prediction error calculation unit 432 moves the reference image up, down, left, and right within the search range. Calculate the absolute value difference for each pixel with the macroblock image data, find the average value with the upper, lower, left, and right pixels, use it as the absolute value difference value with half-pixel accuracy, and find the absolute value difference sum per processing block, The sum of the vector and the absolute value difference at that time is
4 is output.

The search area limiting unit 431 determines whether there is any vector value that extends off the screen within the search range based on the motion vector coarse search result 11 and the macroblock address 21, and determines the effective area within the search range as a prediction error comparison unit. 434. The prediction error comparison unit 434 determines a vector when the sum of absolute value differences per processing block is the smallest in the effective area from the search area restriction unit 431, and outputs the vector to the motion vector output terminal 435.

By controlling the coding mode in this manner, a common circuit realizes frame prediction and field prediction in the frame structure, field prediction in the field structure and 16 × 8 MC, and furthermore, dual prime prediction. Can be done.

Next, the outline of the motion vector search will be described with reference to FIG.

FIG. 6 is a schematic diagram showing an embodiment of a moving picture compression encoding apparatus including a motion vector detecting apparatus according to the present invention. In FIG. 6, the same components as those in FIG. 5 is an image input unit and 6 is a local decoding unit.

The image input from the image input unit 1 is stored in the memory 3. Of the input image, each of macroblock data as a unit for obtaining a motion vector and stored reference image data used for prediction previously input are read from the memory 3 to the motion vector coarse search unit 1. It is.

In this embodiment, the motion vector coarse search unit 1 performs a search with one vertical pixel and two horizontal pixels, and the obtained vector is sent to the motion vector dense search unit 4. The motion vector dense search unit 4 performs a search with a horizontal and vertical 1 pixel precision around the vector from the motion vector coarse search unit 1, and further performs a search with a horizontal and vertical 1/2 pixel precision around the obtained vector. Perform a search.

In order to reduce memory access, the input image data used in the motion vector fine search section 4 is transferred and used for the motion vector coarse search section 1. In the present embodiment, the reference image for the coarse motion vector search unit 1 is an input image, and the reference image for the fine motion vector search unit 4 is a local decoding unit 6 for decoding a locally decoded image obtained by decoding a variable-length encoded stream. Used.

As described above, with less memory access,
By efficiently allocating time slots, it is possible to realize video encoding corresponding to a plurality of prediction modes having a frame structure and a field structure.

In the description of the above embodiment, the sum of absolute difference is used to calculate the prediction error value in the processing block. However, the prediction error value may be calculated using the sum of squares.

In the present embodiment, in order to perform a differential motion vector (DMV) search in dual prime prediction, a different parity (opposite pas- sage) is determined based on a vector obtained with horizontal and vertical one-pixel accuracy in a motion vector dense search.
rity) from the memory,
If there is room in the internal buffer, it is sufficient to read in advance only reference images of different parity before searching for frame prediction, and it is possible to further reduce memory access with a slight increase in buffer capacity.

[0062]

By switching the time slot of the motion vector search, the frame prediction and the field prediction in the frame structure, the field prediction in the field structure and 16 × 8 MC, and even the dual prime prediction can be performed without increasing the circuit scale. Can be realized.

[Brief description of the drawings]

FIG. 1 is a schematic configuration diagram of an embodiment of a motion vector detection device.

FIG. 2 is a diagram showing a processing time slot and a memory access time slot in a B-picture.

FIG. 3 is a diagram showing a P-picture, a processing time slot other than m = 1, and a memory access time slot.

FIG. 4 is a diagram showing a P-picture, a processing time slot in which m = 1, and a memory access time slot.

FIG. 5 is a detailed diagram of a motion vector search in the motion vector detection device.

FIG. 6 is a schematic configuration diagram of an embodiment of a moving image encoding device using a motion vector detecting device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Motion vector coarse search part 2 Encoding control part 3 Memory 4 Motion vector fine search part 5 Image input part 6 Local decoding part 11 Motion vector coarse search result 12 Input image buffer 21 Macro block address 22 Picture type 23 GOP structure 41 Vector switching SW 42 dual prime determination unit 43 motion vector detection unit 44 input image switching SW 45 interpolation image creation unit 46 input image buffer 47 reference image read address generation 48 reference image buffer 49 vector output terminal 431 search area limit unit 432 prediction error calculation unit 433 Coding mode control unit 434 Prediction error comparison unit 435 Motion vector output terminal

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Yukito Tsuboi 292 Yoshida-cho, Totsuka-ku, Yokohama-shi, Kanagawa Prefecture Hitachi, Ltd. Multimedia System Development Division

Claims (5)

[Claims] 1. A main vector and a small difference vector are used, reference from the same parity is performed from the main vector, and reference from a different parity is obtained by adding a small difference vector to a main vector scaled according to an inter-field distance. A motion vector detecting device corresponding to a dual-prime prediction method to be performed, comprising: a search unit that inputs a reference image and an input block image used for prediction, and detects an optimal motion vector within a predetermined search range; The search for the motion vector using the search means is performed twice per one input block image. In the bidirectional prediction coded image, the first search is for the motion vector of forward prediction, and the second search is for the backward vector. The search means is controlled so as to search for a motion vector for prediction. In the forward prediction coded image, the forward prediction is performed for the first time. , A main vector of dual prime prediction is generated from the motion vector obtained in the first search, and the control of the search means is performed in the second search to search for a small difference vector of the dual prime prediction. And a search control unit for performing the search. 2. The motion vector detecting device according to claim 1, further comprising: a coarse search means for performing a coarse search for a plurality of types of motion vectors at a stage preceding the search means, wherein the bidirectional prediction coded image is forward and backward. In each of the directions, a single motion vector can be obtained, and the forward prediction coded image can obtain two types of motion vectors with respect to a forward direction. For the bidirectional prediction coded image, the first search is for a forward prediction motion vector, the second search is for a backward prediction motion vector, and the first prediction is for a forward prediction coded image. Search for the motion vector for forward prediction, generate the main vector for dual prime prediction from the motion vector obtained in the first search, and dual prime for the second When the searched small difference vector dual-prime prediction, the motion vector detection apparatus when the dual-prime prohibited, characterized in that the search for the other vectors output from the coarse search unit. 3. The motion vector detecting device according to claim 1, wherein a search per search of said search means is time-divided into a search in units of integer pixels and a search in units of half pixels. A motion vector detecting device, wherein a vector obtained by the search is used as a main vector for dual prime prediction, and the main vector is scaled, and a reference image for prediction of different parity is read from a memory. 4. The motion vector detecting device according to claim 1, wherein the reference image is read from a memory by expanding only a different parity in advance for dual prime prediction. 5. The motion vector detecting apparatus according to claim 3, wherein when a vertical component of a vector obtained by the search in an integer pixel unit indicates a different parity, the vector is approximated to the same parity close to 0 and the dual prime prediction is performed. A motion vector detecting device, wherein the main vector is a motion vector.