JP3460759B2 - Motion vector detection method and apparatus - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a highly accurate motion vector detecting device used for moving image coding.

[0002]

2. Description of the Related Art A moving image coding apparatus has been attracting attention as a technology that plays a central role in multimedia. inside that,
Many coding devices that meet the standards of H261 and MPEG1 have already been developed and put into practical use. recently,
The MPEG2 standard for encoding higher-definition images has been standardized, and encoding LSIs therefor are being actively developed.

[0003] The key to realizing a compact and economical encoding process of H261, MPEG1, MPEG2, etc. is as follows.
It depends on the operation of the motion vector detection device that calculates from which of the images located before and after the block of the image to be encoded is considered to have moved. That is, the processing amount of the matching process (hereinafter referred to as block matching process), which is a standard method of motion vector detection and which checks the degree of matching between the block to be coded and the preceding and following images, is extremely large, and This is because, in addition to occupying most of the conversion processing, it is necessary to input and output a huge amount of data between the memory that stores the image and the motion vector detection device.

FIG. 4 shows a typical structure of a conventional moving picture coding apparatus.

This device comprises a motion compensation unit 201, D
It is composed of a CT / Q unit 202, a VLC unit 203 and an image memory 204. Motion compensation unit 201
Is the image memory 2 for the encoding target image block.
As shown in FIG. 8, an image block regarded as a source is searched from the search area of the reproduced image to be searched, which is read from 04, and from the source to the encoding target image block found thereby. A block matching unit 210 that detects a displacement amount as a motion vector, an address for reading a search region of a reproduced image from the image memory 204, and an address for reading a reproduced image block of a movement source based on the motion vector are generated. The address generator 211 and the subtracter 212 that calculates the difference between the image block to be encoded and the reproduced image block of the movement source and outputs the result as a prediction error image block. DCT
The / Q unit 202 applies the DCT to the prediction error image block.
DCT converter (hereinafter referred to as DCT) 220 that performs conversion
Then, the conversion result is quantized, and the result is VLC unit.
Quantizer (hereinafter referred to as Q) 222 output to the router 203
And an inverse quantizer (hereinafter referred to as IQ) 223 that inversely quantizes the quantization result and restores the state before quantization, and an inverse DCT transform that reproduces a prediction error image block by applying an inverse DCT transform to the output from the IQ 223. Container (hereinafter referred to as IDCT) 221
And an adder 224 that generates a reproduced image block by adding the reproduced prediction error image block and the reproduced image block of the movement source input from the motion compensation unit 201, and outputs the result to the image memory. The reproduced image block includes an address generator 225 that generates a predetermined address so that the reproduced image block forms a reproduced image in the reproduced image memory.

The VLC unit 203 has a DCT /
The Q unit 202 is a unit for applying variable length coding to the input from the Q unit 202 and then outputting it as a bit stream to the outside.

Further, regarding the specific configuration of the block matching unit 210, CH Hsieand TP Lin, "VLSI
Architecture for Block-Matching Motion EstimationA
lgorithm ", IEEE Trans. Circuits & Syst. for Video
Technol., Vol.2, No.2pp. 169-175 (1997) explains in detail.

A single pixel precision search area, a two pixel precision search area, and a half pixel precision search area described below are shown in FIGS. 6 (a), 6 (b) and 6 (c), respectively. ing.

This conventional encoding apparatus has an advantage that the size of the image memory can be reduced because the image to be held during encoding is limited to the reproduced image. It also has a good reputation for encoding quality. Therefore, various encoding devices based on this encoding method have been realized. On the other hand, in order to reduce the calculation amount of block matching, a half-pixel precision motion vector is re-computed in a narrower search range than the center of the movement source obtained between two-pixel precision images generated by subsampling. A hierarchical motion vector detection device for re-detection has been realized. About this, N. Hayashi
et al .: "A Com-pact Motion Estimator with a Simplif
ied Vector Search Strategy Maintain-ing Encoded Pi
cture Quarity, IEEE 1995 CUSTOM INTEFRATED CIRCUIT
CONFER-ENCE, pp. 409-412 (1995).

FIG. 5 is a diagram showing a configuration of the conventional motion vector detecting device 260, and FIG. 9A is an explanatory diagram of a block matching method by the motion vector detecting device of FIG. This motion vector detecting device includes a block matching unit 101 with 2-pixel precision, a block matching unit with half-pixel precision 102, and address generators 110 and 11.
1, a buffer memory 132, 133, and a half-pixel precision image generator 150, and an external image memory 40,
41 is connected.

FIG. 6 (a) is an original image (locally decoded image) with single pixel accuracy, FIG. 6 (b) is a search area with two pixel accuracy, and FIG.
(C) is a figure which shows a search area of a half pixel precision.

The block to be coded with 2-pixel precision input from the outside is a block matching unit 101 with 2-pixel precision.
Block as an image block of the template, and the block on the search target image having the same coordinates as the coordinates on the encoding target image of the encoding target block which is read from the image memory 41 to the buffer memory 132 in advance by the address generator 111. Block matching is performed with the search area centered at to obtain a motion vector with 2-pixel accuracy. Subsequently, the half-pixel precision block matching unit 102 reads out from the buffer memory 130 the half-pixel precision block matching in which the source of movement obtained from the motion vector with two-pixel precision is the center of the search area. The motion vector with half-pixel precision is obtained by performing the search with the half-pixel precision search area output from the generator (see FIG. 9).
(A)). The half-pixel precision search area at this time is a single-pixel precision search area centered on the movement source specified by the motion vector output from the 2-pixel precision block matching unit 101 by the address generator 110. It is generated by reading from the memory and inputting it to the half-pixel image generator 150 via the buffer memory 133.

[0013]

In the conventional motion vector detecting device described above, as shown in FIG. 9 (a), the movement source obtained with the accuracy of 2 pixels is used as a narrow search range, as shown in FIG. 9 (a).
It is necessary to perform block matching and search again for 48 types of movement source candidates in a search range expanded by at least 1.5 pixels in the surroundings. There is a drawback that the block matching device with half-pixel accuracy is not sufficiently reduced, and the hardware scale is increased, and power consumption is increased. Here, FIG. 9 shows the position to which the origin of the template is applied at the time of matching, and the number of times of matching repeated about the origin of the moving source can be clearly read.

An object of the present invention is to search for single pixel accuracy.
From the external storage device to the internal vector of the motion vector detection device.
Based on the search area read in the far memory
By creating the search area, block pixels with single pixel accuracy are generated.
By adding the
Motion in which while preventing that results in an increase in the data input and output amount, sufficient <br/> large comb the effect of operation of the block matching reduction by layering, to reduce power consumption and hardware scale of the half pixel accuracy block matching unit It is to provide a vector detection device.

[0015]

According to the motion vector detecting method of the present invention, an image to be coded with a plurality of pixel accuracy is held.
Motion vector detection device
Inside the motion vector detection device by reading out the search area with elementary precision
Held in the storage means (first internal storage means), a plurality of pixels seminal
Input the block to be encoded,
From the search area, the input encoding of multiple pixels is performed.
Block that can be regarded as the source of movement by matching with the target block
And the code from the block that can be regarded as the searched source
The motion vector up to the block to be converted with multiple image accuracy
Therefore, the motion vector that holds the locally decoded image with single pixel accuracy is
From the storage means external to the cutle detection device,
Considered as the movement source specified by the motion vector obtained in
Read the search area with single pixel accuracy centered on the block
Storage means inside the motion vector detection device (second internal storage
Method) and enter the block to be encoded with single pixel precision.
Input from the search area with single pixel accuracy.
Moved by matching with the target block of single pixel precision encoding
Search for a block that can be regarded as the source, and consider it as the searched source
Motion vector from the block to be encoded to the block to be encoded
Are stored in the second internal storage means.
From the search area with single pixel accuracy
It can be regarded as the movement source specified by the obtained motion vector.
Generate a half-pixel precision search area centered on the block,
From the search area with half-pixel accuracy, the input single image
It is identified as the source of movement by matching with the block to be encoded with elementary precision.
Search for a block that can be considered and consider it as the source of the search.
Half the motion vector from the lock to the target block
It is characterized in that it is obtained with pixel accuracy.

The motion vector detecting apparatus of the present invention is provided with a plurality of images.
Motion vector that holds the image to be encoded with elementary precision
Search area with multiple pixel accuracy from storage means outside the detector
Memory hand inside motion vector detection device to read and hold
Stage (first internal storage means) and encoding target with multiple pixel accuracy
Input a block, and if it is in the search area with multiple pixel accuracy
And the input block to be encoded with the multi-pixel precision
Search for a block that can be considered
Block to be encoded from the block that can be regarded as the moving source
Means for determining a motion vector to a plurality of pixels precision, single
Motion vector holding a pixel-decoded locally decoded image
Obtained with the accuracy of the multiple pixels from the storage means outside the detection device
Block that can be regarded as the source of movement specified by the motion vector
The search area with single pixel accuracy centered on the clock
The storage means inside the motion vector detecting device (second internal storage)
Memory) and the block to be encoded with single pixel accuracy
Then, from the search area with the single pixel precision, the input
Explore the movement source and can be regarded blocked by matching the encoding target block of single pixel accuracy, regarded as searched moving source
Means for obtaining a motion vector from a block to a block to be encoded with single image accuracy, and the second internal storage means.
From the held search area with single pixel accuracy, the single pixel
Considered as the movement source specified by the motion vector obtained with accuracy.
Means for raw <br/> forming a search area of the half-pixel precision around the block to做, from among the half-pixel precision search region, before
For matching with the input single-pixel precision encoding target block
There is provided means for searching for a block that can be more regarded as a movement source, and obtaining a motion vector from the searched movement source block to the encoding target block with half-pixel accuracy.

[0017]

BEST MODE FOR CARRYING OUT THE INVENTION A single-pixel precision search is inserted between a multiple-pixel precision search and a half-pixel precision search, and a half-pixel precision search is centered around the source specified in the single-pixel precision search. By performing this, the number of movement source candidates can be increased from 48 types to 8 types, and the block matching with single pixel accuracy is increased by 8 times.

FIG. 10 is a basic flowchart of the motion vector detecting method of the present invention.

First, a block having a two-pixel precision including the moving source of the block to be coded is read from the image to be coded (step 1). The source of movement obtained by the reading in step 1 is searched for, and a motion vector with 2-pixel precision from that to the encoding target block is obtained (step 2). Up to this point, the operation is similar to the conventional one.

Next, from the moving source block having a precision of 2 pixels,
The source of movement is confined to a block with single pixel precision, and a motion vector with single pixel precision from that block to the block to be encoded is obtained (step 3). A source block with half-pixel precision is searched from the source block obtained in step 3, and a motion vector with half-pixel precision from that source block to the encoding target block is obtained (step 4).

[0021]

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

FIG. 1 is a block diagram of a first embodiment of a motion vector detecting device to which the motion vector detecting method of the present invention is applied.

The motion vector detecting device 160 includes a two-pixel precision block matching unit 101, a single-pixel precision block matching unit 100, a half-pixel precision block matching unit 102, and address generators 110, 111, and 11.
3, a buffer memory 132, 133, and a half-pixel precision image generator 150.
0 and 41 are connected.

FIG. 9B is an explanatory diagram of the motion vector detecting method of the present invention.

The block to be coded with 2-pixel precision input from the outside is a block matching unit 101 with 2-pixel precision.
7B, and the coordinates on the encoding target image of the encoding target block which are sent from the image memory 41 to the buffer memory 132 by the address generator 111 in advance as a template image block as shown in FIG. 7B. Block matching is performed between the search area centered on the block on the search target image having the same coordinates (○ in FIG. 9B).
Mark) A motion vector with a 2-pixel accuracy is obtained. Then, block matching with single pixel accuracy centered on the search area of the movement source specified by this motion vector (Δ in FIG. 9B)
Is a single pixel precision block matching unit 100,
A motion vector with single pixel accuracy is obtained by performing the process between the block to be encoded of the template as shown in FIG. 7A and the search area read from the buffer memory 133. At this time, the search area is surrounded by the surrounding area 1 centered on the source of movement of 2 pixel accuracy.
In the area where the pixels are widened, a motion vector of two-pixel precision output from the block matching unit 101, and a coordinate of a coding target block in a coding target original image that is separately set,
From the image memory 40 to the buffer memory 133 according to the address generated by the address generator 110 based on the start address of the storage destination of the reproduced image to which the source block belongs.
Read to. Further, the half-pixel precision block matching unit 102 performs the half-pixel precision block matching (x mark in FIG. 9B) in which the source of movement obtained from the single-pixel precision motion vector is the center of the search area. It is performed between the image to be encoded and the half-pixel precision search area output from the half-pixel precision image generator to obtain a half-pixel precision motion vector. Search region of the half pixel precision in this, based on the motion vector output from the block matcher 100 single-pixel accuracy, generated by the address generator 113
The buffer area is formed by expanding the moving source specified by the single-pixel precision motion vector at the specified address by one pixel.
It is generated by reading out from No. 3 and inputting it to the half-pixel image generator 150.

Block matching unit 102 with half-pixel accuracy
, The accuracy of the moving source has risen to the single pixel accuracy because of the block matching with the single pixel accuracy. Therefore, the area of uncertainty is reduced from the area expanded by 1.5 pixels around to the area expanded by 0.5 pixels around, and the number of block matching is reduced from 48 times to 8 times (shown in FIG. 2). Is reduced to. Therefore, normally, the performance of the array of arithmetic units used for block matching with half-pixel accuracy can be reduced to 1/6. This means that the array size can be reduced to 1/6 if the performance of the arithmetic units is the same. On the other hand, the scale of the hardware of the block matching device with single pixel precision that is newly required is such that the region of uncertainty is a region expanded by one pixel around the source of movement with two pixel precision. Since the number of times is as small as 9 times, it is equivalent to the hardware of a block matching unit with reduced half-pixel accuracy. After all, the hardware scale of both block matching units, which occupy most of the hardware scale of the motion vector detecting device, is reduced to about 1/3 of the conventional scale. Here, how the number of times of block matching is calculated for the search area, and the size of the template image block is 2 × 2,
FIG. 2 shows that there are eight types of block matching for a 4 × 4 search area in which the surrounding pixel is expanded by one pixel with respect to the source. FIG. 9 is an explanatory diagram of a conventional block matching method and a block matching method of the present invention. The effect of the present application compared with the conventional method can be read from FIG. 9 showing the position to which the origin of the template is applied at the time of matching. In the conventional method, a block pixel with half-pixel accuracy is used .
In the present application, the number of single-pixel precision is 9 and the number of subsequent half-pixel precision block matching is 8, which is 17 in total, whereas the number of hitting is 48 (the number of x). / 3.

FIG. 3 is a block diagram of a second embodiment of the motion vector detecting device of the present invention.

The difference from the first embodiment of this device is that the motion vector at the output of the block matching unit is not directly input to the address generator, but the microprocessor 30
The point is that they are input to the address generators 114 and 115 via 0. This makes the address generators 114, 115 much simpler than the address generators 110, 113. This is because the microprocessor 300 is made to share the complicated pre-processing for accessing the search area from the source address or the like.
As a result, only the start address and the number of read data are input to the address generator, and the function of the address generator becomes simple by reading the start address, incrementing by the number of data, and outputting.

In the present embodiment, the reduced image expressed as a plural pixel precision in the claims has a 2-pixel precision for the sake of simplicity. (4 pixels consisting of 2 pixels in both vertical and horizontal directions are reduced to 1 pixel) Precision of 3 pixels, precision of 4 pixels, or precision of 1 pixel for vertical, precision of 2 pixels for horizontal, and arbitrary multiple pixel precision. Since the search area is expanded, the effect of the present invention is further enhanced.

[0030]

As described above, according to the present invention, a single-pixel precision search is inserted between a multiple-pixel precision search and a half-pixel precision search, and the source specified by the single-pixel precision search is centered. As a result of performing a half-pixel precision search, the number of movement source candidates is changed from 48 types to 8 types, and even if it is considered that single pixel precision block matching is required 8 times, the calculation after the 2-pixel block matching is performed. The amount is reduced to 1/3, the hardware scale of the block matching device except 2 pixel precision is reduced to about 1/3 of the conventional one, and
The amount of data input / output to / from the image memory by adding the single pixel precision block matching because the single pixel precision search region that was conventionally read is used for the single pixel precision block matching to generate the half pixel search region. The effect is that there is no increase in

[Brief description of drawings]

FIG. 1 is a block diagram of a first embodiment of a motion vector detection device to which a motion vector detection method of the present invention is applied.

FIG. 2 is a diagram showing how to count the number of block matching times.

FIG. 3 is a block diagram of a second embodiment of the motion vector detection device of the present invention.

FIG. 4 is a typical configuration diagram of a conventional moving image encoding device.

FIG. 5 is a block diagram of a motion vector detecting device of a conventional moving image encoding device.

6A is an original image or locally decoded image with single pixel accuracy, FIG. 6B is a diagram showing a search area with two pixel accuracy, and FIG.
FIG. 3 is a diagram showing a search area with half-pixel accuracy.

7A is a diagram showing a template for a search with single pixel precision, and FIG. 7B is a diagram showing a template for a search with two pixel precision.

FIG. 8 is a diagram showing a concept of motion vector search.

9A is an explanatory diagram of a block matching method according to a conventional method, and FIG. 9B is an explanatory diagram of a block matching method according to the present invention.

FIG. 10 is a basic flowchart of the motion vector detecting method of the present invention.

[Explanation of symbols]

40 image memory (for non-reduced images) 41 Image memory (for reduced image) 100 single pixel precision block matching device 101 2-pixel precision block matching device 102 Half-pixel precision block matching device 110 address generator 111 address generator 113 address generator 114 address generator 115 address generator 132 buffer memory 133 buffer memory 150 Half Pixel Precision Image Generator 160 Motion vector detection device 201 Motion compensation unit 202 DCT / Q unit 203 VLC unit (VLC) 204 image memory 210 block matching device 211 Address generator 212 Proto-calculator (-) 220 DCT converter (DCT) 221 Inverse DCT Converter (IDCT) 222 Quantizer (Q) 223 Inverse Quantizer (IQ) 224 Adder (+) 225 address generator 260 motion vector detection device

Front Page Continuation (72) Inventor Toshihiro Minami 1-6 Uchisaiwaicho, Chiyoda-ku, Tokyo Nippon Telegraph and Telephone Corporation (72) Naoki Ono 1-1-6 Uchisaiwaicho, Chiyoda-ku, Tokyo Nippon Telegraph and Telephone Corporation In-house (56) Reference JP-A-5-284489 (JP, A) JP-A-4-23594 (JP, A) JP-A-4-288789 (JP, A) N.P. Hayasi et al. : "A Compact Motion Estimator with a S implified Vector S earch Strategy Mai ntaining Encoded P icture Quality, IEE E 1995 CUSTOM INTEGR ATED CIRCUIT CONFE RENCE, 409-412 (1995) (58) investigated the field (Int.Cl. ^7, DB name) H04N 7/137

Claims (2)

(57) [Claims] 1. An image to be encoded having a multi-pixel precision is retained.
The motion vector detection device is
A motion vector detection device that reads out a search area with several pixel accuracy
The plurality of pixels are stored in an internal storage unit (first internal storage unit) and a block to be encoded with a plurality of pixel accuracy is input.
From the pixel-precision search area, the input multiple pixel
Deemed to be the source of movement by matching with the block to be encoded
Block that can be regarded as the source of the search.
Multiple motion vectors from the block to the block to be encoded.
Motion vector obtained with image accuracy and holding a locally decoded image with single pixel accuracy
From the storage means external to the module
The motion source specified by the motion vector
Reads a single-pixel precision search area centered on the lock
Storage means inside the vector detector (second internal storage
Row), input the block to be encoded with single pixel precision, and input the single pixel
From the precision search area, input the single pixel precision code
A block that can be regarded as the source of movement by matching with the block to be encoded.
From a block that can be regarded as the source of the search.
Motion vector up to the block to be encoded with single image accuracy
The single pixel precision search held in the second internal storage means is obtained.
From the search area, the motion vector obtained with the single pixel accuracy is used.
Half centered on a block that can be regarded as a designated source
A search area with pixel accuracy is generated, and the input single image is selected from the search area with half-pixel accuracy.
It is identified as the source of movement by matching with the block to be encoded with elementary precision.
Search for a block that can be considered and consider it as the source of the search.
Half the motion vector from the lock to the target block
A motion vector detection method characterized by being obtained with pixel accuracy . 2. An image to be coded with multiple pixel accuracy is retained.
The motion vector detection device is
A motion vector that reads and holds a search area with several pixel accuracy
A storage means (first internal storage means) inside the detection device and a plurality of pixel-precision encoding target blocks are input,
From the pixel-precision search area, the input multiple pixel
Deemed to be the source of movement by matching with the block to be encoded
Block that can be regarded as the source of the search.
Multiple image motion vector to the encoding Target block from the click
A method for obtaining with precision and a motion vector that holds a locally decoded image with single pixel precision
From the storage means external to the module
The motion source specified by the motion vector
The search area with a single pixel accuracy centered on the lock is read out and stored.
The storage means inside the motion vector detecting device (the second inside)
(Storage means) and a block to be encoded with single pixel accuracy are input, and the single pixel
From the precision search area, input the single pixel precision code
A means for searching a block that can be regarded as a movement source by matching with the encoding target block, and obtaining a motion vector from the block that can be regarded as the movement source searched to the encoding target block with single image accuracy, The single pixel precision search held in the second internal storage means.
From the search area, the motion vector obtained with the single pixel accuracy is used.
Half centered on a block that can be regarded as a designated source
Means for generating a search area of the pixel accuracy, from among the half-pixel precision search region, Tan'e that the input
A block that can be regarded as the source of movement is searched by matching with the block to be encoded with elementary precision, and a block that can be regarded as the searched source of movement is searched.
A motion vector detection device having means for obtaining a motion vector from a lock to a block to be encoded with half-pixel accuracy.