JP3437605B2 - Image motion 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 an image motion detecting method and device mainly used for compressing moving images.

[0002]

2. Description of the Related Art As one of the moving image compression techniques, there is a hybrid system in which motion compensation and discrete cosine transform (DCT) are combined. Two
It is also adopted in 61. Here, in the motion compensation, as shown in FIG. 8, the image (a) of the past frame is more faithfully combined with the image (b) of the current frame, and the prediction error of the motion-compensated image combined with the image of the current frame is calculated. Used to reduce

A block matching method is generally used for motion compensation. In the block matching method, the image to be encoded is first divided into blocks of a square lattice, and one representative point is placed at the center of the block. Next, a block in which the mean square error for each pixel is the smallest is searched from the range of 16 × 16 pixels of the past frame for the block of interest, and the distance between the block of interest and the searched block is calculated as the representative point motion vector. To detect as. After detecting the representative point motion vector 2 for the representative point of each block, the block 1 indicated by the representative point motion vector 2 from the past frame
Is cut out and used as the motion compensation image of the block 3.

In this block matching method, a motion-compensated image is cut out in units of blocks having a predetermined area, so that block distortion appears. Cutting out a block is equivalent to setting the motion vector for each pixel (referred to as a pixel motion vector) to be the same as the representative point motion vector for all pixels in one block. For example, in FIG.
When the representative point motion vector V is given to the representative point Q at the center of the block B of 4 × 4 pixels, as shown in, the pixels P11 to P44 in the block B are the same pixels as the representative point motion vector V. Motion vectors V11 to V14 are given, and the pixel values of the past frame pointed to by each are clipped into blocks as shown in FIG. 8 to be the respective predicted pixels.

As described above, in the block matching method , all the pixel motion vectors in the block are equal to the representative point motion vector. Therefore, for example, when the pixel motion vector of the original image continuously changes as shown in FIG. No motion vector can be obtained. Also, when the block includes the edge E where the pixel motion vector is discontinuous as shown in FIG. 11, the correct pixel motion vector cannot be obtained. Therefore, the prediction error becomes large, and the movement of the original image cannot be represented correctly.

On the other hand, in the document "Basic Study of Motion Compensation Method at Ultra-Low Rate" (Nakaya, Harashima et al., Autumn Society of 1992, Shinzen University, D-145), a new motion compensation method replacing the block matching method. As a result, we are considering a motion compensation method based on a triangular patch. In this method, a triangular patch is attached to the frame to be encoded, and the motion vector of the vertex of the triangle is detected as the representative point motion vector. In this method, a pixel motion vector of a pixel inside a triangle is obtained from these representative point motion vectors by texture mapping based on affine transformation, and a motion compensation image is synthesized.

Also, a method of sticking square patches has been proposed (GJ Sullivan, RLBaker, “Motion com.
pensation for video compression using control grid
interpolation, "ICASSP 91, pp. 2713). As an example, consider the case of using the quadrilateral patch motion compensation having an area of 4 × 4 pixels shown in FIG. 12, and the vertices R1, R2, R of the quadrilateral patch R.
3, R4 motion vector, that is, representative point motion vector V
It is assumed that 1, V2, V3 and V4 are detected. At this time, the pixel motion vector of the pixel included in the square patch R is obtained by bilinear interpolation of the four representative point motion vectors V1, V2, V3, V4 as shown in FIG.

In such a so-called patch motion compensation system, it is possible to compensate for the drawback of the block matching method that cannot deal with the occurrence of block distortion and continuous change of motion vector. However, for example, in the original image of the frame to be encoded as shown in FIG. 14, even when the square patch R includes an edge E ′ having a discontinuous pixel motion vector, FIG.
Since the pixel motion vectors are continuously combined as in the case of 3, the disadvantage that the discontinuous motion of the image cannot be expressed still remains.

[0009]

As described above, the conventional block matching method and patch motion compensation method cannot correctly detect a pixel motion vector when a block includes discontinuous edges in the pixel motion vector. There was a problem.

It is an object of the present invention to provide a motion detection method and device capable of correctly detecting discontinuous pixel motion vectors.

[0011]

In order to solve the above-mentioned problems, the present invention divides a motion detection target image into a plurality of blocks composed of a plurality of pixels, and represents each block with a reference image. In a motion detection method of detecting a point motion vector and further synthesizing a pixel motion vector indicating the motion of each pixel, in each block of the motion detection target image, a plurality of pixels indicated by the representative point motion vector from the reference image A motion-compensated image is generated by cutting out a block composed of motion-compensated images, and an average error of pixel values in a block between the generated motion-compensated image and the motion-detection target image is calculated. It is determined whether the obtained error is less than or equal to a predetermined threshold value, and as a result of the determination, if the error is greater than the threshold value, the motion detection target image The motion compensation is performed by switching the geometrical shapes of blocks of the lock and the reference image and the block shapes including the area and the position, and as a result of the determination, when the error is equal to or less than the threshold value, A block shape is determined, and the pixel motion vector in the block is synthesized using the representative motion vector of the block in which the block shape of the motion detection target image is determined.
In another motion detection method according to the present invention, for each block of the motion detection target image, a block composed of a plurality of pixels indicated by the representative point motion vector is cut out from the reference image to perform motion compensation. A motion compensation image is generated, and an average error of pixel values in a block between the generated motion compensation image and the motion detection target image is calculated,
The error is minimized by switching the geometrical shape of the block of the motion detection target image and the block of the reference image, and switching the block shape including the area and position a prescribed number of times and performing the motion compensation each time the block shape is switched. The block shape is determined, and the pixel motion vector in the block is synthesized using the representative motion vector of the block of the motion detection target image in which the block shape is determined.

The motion detection apparatus according to the present invention divides the motion detection target image into a plurality of blocks each composed of a plurality of pixels, detects a representative point motion vector for each block between the reference image and the pixel, and further detects the pixel In a motion detection device for synthesizing a pixel motion vector indicating each motion, for each block of the motion detection target image, a block composed of a plurality of pixels indicated by the representative point motion vector is cut out from the reference image to perform motion. A motion compensation unit that generates a motion compensation image by performing compensation, and an average error of pixel values in a block between the motion compensation image generated by the motion compensation unit and the motion detection target image is detected. Error detection means, determination means for determining whether or not the error detected by the error detection means is less than or equal to a predetermined threshold value, and the determination hand If the error is larger than the threshold value as a result of the determination by the block shape switching means, the block shape switching means switches the geometric shape and the block shape including the area and the position of the block of the motion detection target image and the block of the reference image. However, every time the block shape is switched by the block shape switching means, the motion compensation by the motion compensating means, the error detection by the error detecting means, and the determination by the determining means are performed. Is less than or equal to the threshold value, the block shape is determined at that point, and the pixel in the block is determined using the representative motion vector of the block in which the block shape of the motion detection target image is determined. Synthesize motion vectors. According to another motion detection device of the present invention, for each block of the motion detection target image, a block composed of a plurality of pixels indicated by the representative point motion vector is cut out from the reference image to perform motion compensation. Motion compensation means for generating a motion compensation image; error detection means for detecting an average error of pixel values in a block between the motion compensation image generated by the motion compensation means and the motion detection target image; Block shape switching means for switching the block shape including the geometrical shape and area and position of the block of the motion detection target image and the block of the reference image a prescribed number of times, and each time the block shape is switched by the block shape switching means. , The motion compensation means performs the motion compensation and the error detection means detects the error, and the error is minimized. Determine the Kunar said block-shaped, combining the pixel motion vector in the block using the representative motion vector of the block to block shape is determined in the motion detection target image.

[0013]

[Operation] A block between the motion compensation image and the image to be encoded
When the average error of the pixel value in the pixel is detected, the error becomes large if the block includes a discontinuous contour. Therefore, a block shape that reduces this error
If decision, discontinuous contour is to be excluded from the internal block, the pixel motion vector is correctly detected.

[0014]

As described above, according to the present invention, by changing the block shape when performing adaptive motion compensation, a portion in which the motion vector continuously changes is continuously synthesized, and a contour portion in which the motion is discontinuous is formed. On the other hand, by performing image synthesis similar to actual motion, it is possible to reduce deterioration of the motion compensation image.

[0016]

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

FIG. 1 is a block diagram of a moving picture coding apparatus to which a motion detecting apparatus according to the present invention is applied. In the figure, the encoding target image signal 10 which is a moving image signal is input to the subtractor 11 and the difference from the motion compensated image signal 20 which is a predicted image signal is obtained to generate a prediction residual signal 13. . The prediction residual signal 13 is discrete cosine transformed by a DCT (discrete cosine transform) circuit 14. The DCT coefficient data obtained by the DCT circuit 14 is transferred to the quantization circuit 15
Is quantized by. The signal quantized by the quantization circuit 15 is branched into two, one of which is variable-length coded by the variable-length coding circuit 16 and, after passing through a buffer (not shown) to a predetermined transmission rate, a transmission line or a storage system. Output to media.

On the other hand, the other of the signals quantized by the quantizing circuit 15 and bifurcated, the other is the inverse quantizing circuit 17 and the inverse DCT.
The circuit 18 sequentially performs the processing reverse to the processing in the quantization circuit 15 and the DCT circuit 14, and then is added to the motion compensation image signal 20 in the adder 19 to generate the local decoded signal 21.

The locally decoded signal 21 thus generated is
It is input to the motion compensation circuit 23 via the frame memory 22. The motion compensation circuit 23 includes a motion vector detection circuit 24.
By inputting the motion vector (representative point motion vector) from the frame, the local decoded signal read from the frame memory 22 is used as a reference image signal to perform motion compensation on this reference image signal , and the motion compensated image signal 20 is obtained. Output. The motion vector detection circuit 24 detects a motion between frames using the image signal 10 to be encoded and the output signal of the frame memory 22, and generates a representative point motion vector which is information indicating the direction and amount of the motion. The motion compensation image signal is generated in the motion compensation circuit 23 according to the representative point motion vector and the pixel motion vector generated in the motion compensation circuit 23.

The error detecting circuit 25, the judging circuit 26 and the block shape switching section 27 together with the motion compensating circuit 23 and the motion vector detecting circuit 24 constitute a motion detecting device according to the present invention. The error detection circuit 25 is a block internal image between the encoding target image signal 10 and the motion compensation image signal 20.
An average error of the elementary values, for example, a root mean square error described later is detected. The determination circuit 26 determines the magnitude of the error detected by the error detection circuit 25, and controls the block shape switching unit 27 based on the determination. The block shape switching unit 27
The block shape in which the motion vector detection circuit 24 performs pixel motion vector synthesis , that is, the motion compensation circuit 23 performs motion compensation.
The block shape is changed when performing .

Next, the motion detecting operation in this embodiment will be described with reference to FIGS. FIG. 2 is a conceptual diagram of the motion detection operation, and FIG. 3 is a diagram showing an example of the algorithm.

First, in the motion vector detection circuit 24, the image to be encoded is divided into a plurality of blocks B having a predetermined shape.
In FIG. 2, the blocks are arranged in a square lattice. For each of these blocks B, one representative point is set in the center (not shown). Each divided block B
On the other hand, in the range of 16 × 16 pixels of the image of the past frame from the frame memory 22, a block in which the average of the squared error for each pixel is the smallest is searched, and the distance to the block is calculated as the representative point motion. The motion vector detection circuit 24 detects the vector. Then, the motion compensation circuit 23 performs motion compensation in block units using the detected representative point motion vector to obtain the motion-compensated image signal 20 (S).
11).

Next, the mean square error for each pixel in the block between the motion-compensated image signal 20 thus obtained and the image signal 10 to be encoded is obtained (S12). This mean square error is compared with a predetermined threshold value in the judgment circuit 26 (S13), and if it is not less than the threshold value, the block shape switching unit 27 switches the block shape (S14), S11.
~ The process of S12 is repeated.

As a result, if the mean square error is less than or equal to the threshold value, the block shape at that time is determined (S15), and the pixel motion within the block is determined using the representative point motion vector of the block for which the shape is determined. Combine vectors (S1
6). As a pixel motion vector synthesizing method, a pattern matching method used in conventional motion compensation, affine transformation, bilinear interpolation, or the like can be considered. Here, the case of using bilinear interpolation will be described.

Here, as a method of changing the block shape, for example, when there is a discontinuous contour line E in the image as shown in FIG. By moving the representative point P1 corresponding to the upper left apex of the block B on which the linear interpolation is performed, it is changed to a block B'represented by quadrangular vertices, that is, representative points P1 ', P2', P3 'and P4'. In this way, each vertex of the new block B ′ whose block shape has changed, that is, the representative point P1 ′,
Representative point motion vector V1 'of P2', P3 ', P4',
For V2 ', V3', V4 ', the representative point motion vectors V1, V2, V3, V4 of the vertices P1, P2, P3, P4 of the block B before the block shape change are used as they are.

As shown in FIG. 2 (a), when the block B'after the block shape change becomes smaller than the block B before the change, the difference between the two blocks, that is, the area excluding the block B'from the block B is detected. Pixel (P11 in the example of FIG. 2)
~ P14, P21, P31, P41) pixel motion vector V11 ~
V14, V21, V31, and V41 are set to 0 and determined by interpolating other blocks around. If the mean square error for each pixel of the changed block B'becomes less than or equal to the threshold due to such a change in block shape, the shape of the block to be interpolated is determined to be B ', and if it is less than or equal to the threshold. If not, switch the block shape again.

According to this algorithm, the mean square error for each pixel can be made smaller when the discontinuous contour E is outside the block in which the pixel motion vector is interpolated. Therefore, the pixel motion vectors V11 to V14, V21, V31, and V41 of the pixels P11 to P14, P21, P31, and P41 at the difference between the blocks B'and B after the block shape change are the block B after the change. Representative point motion vectors V1 ', V2', V3 ',
The pixel motion vector Vij is removed from the block B for performing bilinear interpolation by V4 ', and is synthesized as shown in FIG. As a result, a pixel motion vector closer to the image to be coded is obtained as compared with the pixel motion vector obtained by the block matching method as shown in FIG. 9 and the pixel motion vector obtained by the patch motion compensation as shown in FIG. The rate of change of the block shape may be determined by, for example, obtaining an average prediction error of the motion compensation image with respect to the encoding target image and determining whether or not this is less than or equal to the prediction error of the composite image before the block shape change.

From the pixel motion vector synthesized by bilinear interpolation, the pixel value in the past frame pointed to by the pixel motion vector is set as the value of the pixel of interest of the image to be coded. If the pixel position of the other frame pointed to by the pixel motion vector is not an integer,
Pixel values are determined by bilinear interpolation. Here, the other frame is the same as the frame used when detecting the representative point motion vector. Then, by performing the above procedure for each block in the frame of the image to be encoded, the motion-compensated image of the encoded frame is synthesized from another frame.

The block shape may be a polygon having a predetermined area, but here, for simplification, a square of 4 × 4 pixels is considered. The setting position of the representative point may be anywhere in the block. This may be obtained from past frames or future frames. However, in the case of obtaining the set position of the representative point from the future frame, when the moving image compression is performed using the motion compensation according to the present invention, the decoder side must decode the future frame first.

As a method of changing the block shape,
A method of shifting one vertex of the block or a method of shifting all vertices may be used. It may also be achieved by deforming the block into another polygon. It is also possible to change the geometrical shape of the block without changing the area of the block .

When the block B'after the block shape change is smaller than the block B before the change, the pixel motion vector of the difference portion between the two blocks is determined from the motion vectors of surrounding pixels and representative points. You may use the method of inserting. When the block B ′ after the block shape change is larger than the block B before the change, both blocks partially overlap, but the pixel motion vector of this overlapping part is the pixel motion vector obtained after the change. It may be used, or a weighted average with the pixel motion vector before change may be taken.

If the pixel position of the other frame pointed to by the pixel motion vector is not an integer, the pixel position may be approximated by some method such as rounding off the decimal part.

By repeating the above procedure a plurality of times for one frame, it is possible to obtain a motion-compensated image that is closer to the original image. At that time, the detection method of the motion vector detection circuit 24 for detecting the representative point motion vector can be changed according to the change in the block shape. For example,
When the representative point motion vector is detected by pattern matching, it is conceivable to match the shape of the block for pattern matching with the block having the changed shape for motion compensation.

The algorithm of FIG. 3 can be modified as shown in FIG. In FIG. 4, S21, S22, S
Steps S24, S26 are S11, S12, S of FIG.
14 and S16. In this algorithm, the block shape is changed up to a predetermined number of times (S23
Up to S24), the block shape having the smallest mean square error obtained in S22 is determined from among these block shapes (S25).

It is also possible to combine the algorithms of FIGS. 3 and 4 as another example. That is, the algorithm of FIG. 3 is basically used to determine the block shape when the mean square error becomes equal to or smaller than the threshold value. However, even if the block shape is changed a prescribed number of times, the mean square error must be equal to or smaller than the threshold value. According to the algorithm of FIG. 4, the block shape having the smallest mean square error is determined.

The effects of this embodiment are shown below.

Image 1 Image 2 Block matching Conventional technology 37.0 dB 33.6 dB This embodiment 37.4 dB 33.8 dB Motion vector interpolation Conventional technology 38.4 dB 34.5 dB This embodiment 38.7 dB 34.8 dB In both the case where the block matching method is used for the pixel motion vector detection and the case where the motion vector interpolation method is used, the SN of the image is larger in this embodiment.
It can be confirmed that R is improved.

Next, another embodiment of the present invention will be described. FIG. 5 is a functional block diagram showing a configuration of a main part in the embodiment, and shows a portion related to the present invention in the motion compensation circuit and the motion vector detection circuit. In the present embodiment, a representative point representing a block of a predetermined shape is arranged and set at an appropriate position with respect to the pattern or movement of the frame of the image to be encoded, and a motion detecting device that gives a plurality of motion vectors to the representative point is used. is there.

In FIG. 5, the image signal to be encoded is input to the contour detection circuit 31, and the contour E1, as shown in FIG.
E2 is detected. As a contour detection method, for example, there is a method of detecting a portion where a change in luminance between pixels is larger than a predetermined value as a contour. The detection result of the contour detection circuit 31 is input to the representative point setting unit 32, and the representative point Pi is set on the contour. Information on the set representative point is input to the interpolation area determination unit 33 and the representative point motion vector detection unit 34. The representative point motion vector detected by the representative point motion vector detection unit 34 is also input to the pixel motion vector interpolation unit 35.

The representative point motion vector detection unit 34 obtains, as a representative point motion vector, a motion vector representing a region having a contour as a boundary, that is, a region having a representative point as an end point,
When it is determined that they have the same motion vector with the contour as a boundary, the contour is not detected as a discontinuous contour. As the motion vector used for this determination, for example, a representative point is virtually set in each area and the representative point motion vector is used.

With respect to the contour and the image synthesizing method thus detected, representative points are set so that a motion-compensated image close to the original image can be synthesized. Here, a patch motion compensation method will be described as an image synthesizing method. The representative points can be set by setting both ends of the detected line segment of the detected contour, setting representative points at predetermined intervals on the contour, and discontinuity of movement especially in order to reduce the number of representative points. There is a method of extracting a different contour and setting a representative point on such a contour with emphasis. A representative point may be set for each object that makes the same movement, a motion vector may be obtained for each object that makes the same movement, and this may be given to each representative point.

The interpolation area determining unit 33 divides the image to be coded into polygonal patches based on the representative points set in this way, for example, to interpolate pixel motion vectors. Determine the area. As a method of dividing the patch, for example, a method of connecting the representative points having the closest distance between the representative points and dividing the representative points is used.

With respect to the interpolation area thus determined, the pixel motion vector interpolation unit 35 interpolates the pixel motion vector using the representative point motion vector detected by the representative point motion vector detection unit 34. . As this interpolation method, bilinear interpolation is used in the case of square patch motion compensation. At this time, it is determined which of the regions I and J in FIG. 6 in which the pixel for which the pixel motion vector is to be interpolated belongs to has an edge as a boundary, and which region the representative point motion vector corresponding to is used is determined. By this procedure, a discontinuous boundary can be expressed along the edges E1 and E2 as shown in FIG. 6, and the pixel motion vectors of continuous areas can be combined.

Various methods of detecting the contour in the contour detecting section 31 can be considered. There is also a method of independently detecting two types of contours of the image to be encoded in the x direction and the y direction, or individually detecting a plurality of contours in multiple directions. Further, the contour may be detected from the difference image from the previous frame, or may be detected from the change of the pixel value in the frome. In this case, the background area is used to predict the still area. The detected contours may be integrated and simplified by integrating with the neighboring contours. Further, it is possible to simplify the contour detection by pre-sampling the image in advance before detecting the contour.

As the image synthesizing (interpolating) method in this embodiment, a block matching method and a patch motion compensation method can be considered.

There are several other possible methods for setting the representative point. For example, as shown in FIG.
Representative points Pj and Pi are set for the areas J and I, and individual representative point motion vectors Vj and Pi are set in the respective areas.
When Vi is given, in order to synthesize the image while expressing the contour detected by the above method, the positions (x
j, yj), (xi, yi), respective motion vectors Vj, Vi, feature point position information (xi ^E , yj ^E ) of a boundary line approximated by the image contour, and patch formation information are not sent. must not.

However, when the representative points Pj and Pi are arranged on the contour line as shown in FIG. 7 and two types of representative point motion vectors Vj and Vi for each area are given as in this embodiment, it is necessary. Information is the position of the representative points Pj, Pi (x
j, yj), (xi, yi) and the representative point motion vector V
Only j and Vi are required. This is because the position information of the representative point includes the contour line information. The patch formation information can be sent, but is unnecessary according to the method described below.

There are several possible methods of dividing an image into patches to be interpolated from the arranged representative points, in addition to the above method. If it is a triangle patch, the existing Voronoi division method (FPPreparata, MIShamos COMPUTATIONAL GEOMETR
Y, 5.5) can also be used to divide the patch into more equilateral triangles. This polygon division method can also be used for polygons. Not only can the division criterion of the triangle division be closer to an equilateral triangle, but the pixel motion vector can also be divided so as to interpolate with less prediction error. Further, it is also possible to divide by using the pixel information of the previous frame. When dividing into patches using the triangle division method, if the periphery of the moving object is stationary at the center of the screen, it is possible to divide it by giving some virtual representative points outside the screen. It is possible to divide the moving area and the still area by sending only one motion vector in the patch formed using the representative points.

[0049]

As described above, according to the present invention, when a block includes pixel edges having discontinuous pixel motion vectors, pixel motion that cannot be detected correctly by the conventional block matching method or patch motion compensation method. It is possible to reliably detect a vector while performing image synthesis for continuous motion, and it is possible to obtain a motion-compensated image with little deterioration.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a moving picture coding device to which a motion detection device according to an embodiment of the present invention is applied.

FIG. 2 is a conceptual diagram of a motion detection operation of the same embodiment.

FIG. 3 is a flowchart showing an example of a motion detection algorithm of the same embodiment.

FIG. 4 is a flowchart showing another example of the motion detection algorithm of the same embodiment.

FIG. 5 is a block diagram showing a configuration of a main part of a motion detection device according to another embodiment of the present invention.

FIG. 6 is a conceptual diagram of a motion detection operation of the same embodiment.

FIG. 7 is a diagram showing an example of arrangement of representative points along a conventional contour.

FIG. 8 is a principle diagram of motion compensation in the block matching method.

FIG. 9 is a diagram showing an example of pixel motion vector detection by a block matching method.

FIG. 10 is a diagram showing an example in which changes in pixel motion vectors are continuous within a block.

FIG. 11 is a diagram showing an example in which changes in pixel motion vector are discontinuous within a block.

FIG. 12 is a diagram showing an example of representative points and representative point vectors for square patch motion compensation.

FIG. 13 is a diagram showing an example of a pixel motion vector when interpolation is performed by a square patch motion compensation method.

FIG. 14 is a diagram showing an example in which changes in pixel motion vector are discontinuous within a patch.

[Explanation of symbols]

10 ... Encoding target image signal 13 ... Prediction error signal 14 ... DCT circuit 15 ... Quantization circuit 16 ... Variable length coding circuit 17 ... Inverse quantization circuit 18 ... Inverse DCT circuit 20 ... Motion compensation image signal 21 ... Local decoded signal 22 ... Frame memory 23 ... Motion compensation circuit 24 ... Motion vector detection circuit 25 ... Error detection circuit 26 ... Judgment circuit 27 ... Block shape switching unit 31 ... Contour detection unit 32 ... Representative point setting unit 33 ... Interpolation area determination unit 34 ... Representative point motion vector detection unit 35 ... Pixel motion vector interpolation unit

Continuation of the front page (56) References JP-A-1-179584 (JP, A) JP-A-2-13079 (JP, A) Kazuhide Kato (four others), motion compensation method using equal density area , Image Coding Symposium (PCSJ92) 7th Symposium Material, Japan, The Institute of Electronics, Information and Communication Engineers, October 7, 1992, p. 225-226 (58) Fields investigated (Int.Cl. ⁷ , DB name) H04N ^7/ 24-7/68

Claims (4)

(57) [Claims] 1. A motion detection target image is divided into a plurality of blocks composed of a plurality of pixels, a representative point motion vector for each block is detected with respect to a reference image, and a pixel indicating a motion for each pixel is further detected. In a motion detection method of synthesizing motion vectors, for each block of the motion detection target image, motion compensation is performed by cutting out a block composed of a plurality of pixels indicated by the representative point motion vector from the reference image. A compensation image is generated, an average error of pixel values in a block between the generated motion compensation image and the motion detection target image is calculated, and whether the calculated error is less than or equal to a predetermined threshold value. If the result of the determination is that the error is larger than the threshold value, the geometric shape and area of the block of the motion detection target image and the block of the reference image are determined. And the block shape including the position is switched to perform the motion compensation, and as a result of the determination, when the average error is equal to or less than the threshold value, the block shape is determined at that time, and the motion detection is performed. A motion detection method for synthesizing the pixel motion vector in the block using the representative motion vector of the block in which the block shape of the target image is determined. 2. A pixel for which a motion detection target image is divided into a plurality of blocks composed of a plurality of pixels, a representative point motion vector for each block is detected with respect to a reference image, and a pixel showing a motion for each pixel is detected. In a motion detection method of synthesizing motion vectors, for each block of the motion detection target image, motion compensation is performed by cutting out a block composed of a plurality of pixels indicated by the representative point motion vector from the reference image. A compensation image is generated, an average error of pixel values in a block between the generated motion compensation image and the motion detection target image is calculated, and the geometry of the block of the motion detection target image and the block of the reference image By switching the block shape including the shape, area and position a prescribed number of times and performing the motion compensation each time the block shape is switched, the error is reduced. Motion detection method also the block-shaped determined to be smaller, combining the pixel motion vector of the representative motion vector using the said block of said block block shape is determined in the motion detection target image. 3. A motion detection target image is divided into a plurality of blocks composed of a plurality of pixels, a representative point motion vector for each block is detected with respect to a reference image, and a pixel indicating a motion for each pixel is further detected. In a motion detection device for synthesizing motion vectors, for each block of the motion detection target image, motion is performed by cutting out a block composed of a plurality of pixels indicated by the representative point motion vector from the reference image and performing motion compensation. Motion compensation means for generating a compensation image; error detection means for detecting an average error of pixel values in a block between the motion compensation image generated by the motion compensation means and the motion detection target image; Determination means for determining whether or not the error detected by the error detection means is less than or equal to a predetermined threshold value; When the difference is larger than the threshold value, the block shape switching means for switching the block shape including the geometric shape and the area and the position of the block of the motion detection target image and the block of the reference image, the block shape switching Each time the block shape is switched by the means, the motion compensation by the motion compensation means, the error detection by the error detection means and the determination by the determination means are performed, and as a result of the determination, the error is equal to or less than the threshold value. In such a case, the block shape is determined at that time, and the motion for synthesizing the pixel motion vector in the block using the representative motion vector of the block in which the block shape of the motion detection target image is determined. Detection device. 4. A pixel in which a motion detection target image is divided into a plurality of blocks composed of a plurality of pixels, a representative point motion vector for each block is detected with respect to a reference image, and a pixel indicating a motion for each pixel is detected. In a motion detection device for synthesizing motion vectors, for each block of the motion detection target image, motion is performed by cutting out a block composed of a plurality of pixels indicated by the representative point motion vector from the reference image and performing motion compensation. Motion compensation means for generating a compensation image; error detection means for detecting an average error of pixel values in a block between the motion compensation image generated by the motion compensation means and the motion detection target image; Block shape switching for switching the block shape including the geometrical shape and area and position of the block of the motion detection target image and the block of the reference image a specified number of times A block that minimizes the error by performing the motion compensation by the motion compensation means and the error detection by the error detection means each time the block shape is switched by the block shape switching means. A motion detection device that determines a shape and synthesizes the pixel motion vector in the block by using the representative motion vector of the block in which the block shape of the motion detection target image is determined.