JPH05130588A - Motion vector detection device for image - Google Patents

Abstract

PURPOSE:To detect motion vectors between fields which are vertically out of phase with high reliability without causing any large increase in throughput due to repetitive arithmetic, etc. CONSTITUTION:This motion vector detection device which employs a block matching method decides the attributes of respective blocks of an input image signal by an attribute decision circuit 11 by using at least the ratio of contour parts. A block size determining circuit 12 determines block size for motion vector detection according to the decision result so that the block size is larger and larger as the ratio of the contour parts is larger and larger. A motion vector detecting circuit 14 detects motion vectors by using the block size.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image motion vector detecting device applied to an image signal such as a TV signal.

[0002]

2. Description of the Related Art A motion vector detecting device is used, for example, when performing motion compensation predictive coding in a moving picture coding device. In motion compensation predictive coding, a prediction signal is generated by locally moving a reference image based on a motion vector of an input image with respect to a reference image, and a prediction error signal that is a difference between this prediction signal and the input image. Is encoded by an orthogonal transformation such as DCT (discrete cosine transformation). As the reference image, a past image, a future image, or both images are used for the input image.

The motion vector is generally detected by the block matching method. A conventional method of detecting a motion vector by block matching will be described with reference to FIG.

In FIG. 6, a subject represented by a solid line triangle moves between frames (or fields) that are temporally different. Note that the triangular and rectangular blocks represented by broken lines are in one frame (or field).
The subject and the block above are projected on the other frame (or field). The motion vector is defined by the moving direction and the moving amount between the objects, which are represented by the solid line and broken line triangles.

The conventional motion vector detection by the block matching method is performed in the following procedure. First, a block of a certain size is cut out from the input image signal, and at the position indicated by the motion vector within the motion vector search range from the image signal of the previous frame (previous field) stored in the frame (field) memory. Then, a block having the same size as the block of the input image signal is cut out, and the difference value between these blocks is obtained. And
This difference value is evaluated for each block by a predetermined evaluation value, for example, the sum of absolute values or the sum of squares, and the vector having the smallest evaluation value is selected as the final motion vector.

In this conventional motion vector detection method, for example, in a block including a relatively flat portion of an image, there is a vector whose evaluation value is smaller than the correct motion vector due to the influence of noise contained in the image signal. Therefore, the motion vector may not reflect the motion of the subject in the image. Japanese Unexamined Patent Publication No. 1-176178 discloses a motion vector detection method capable of eliminating such a false detection of a motion vector.

In this improved motion vector detecting method, first, the reliability of each motion vector obtained by the motion vector detecting circuit is judged for each block. As a result of this determination, for a block with low reliability, the block size is expanded by integrating the block and the peripheral blocks, and motion vector detection is newly performed for this expanded block. Then, the block size expansion and the motion vector detection are repeated until a highly reliable motion vector is obtained. That is, since the noise included in the image signal does not have inter-pixel correlation, when the block size is expanded, the increase amount of noise power is not proportional to the block size, and the signal component greatly increases. It takes advantage of the fact that the effect of noise is relatively small.

The motion vector obtained by this method well represents the global motion of the subject in the image, and erroneous detection of the motion vector is significantly reduced as compared with the block matching method described above. However, in this method, since the determination of the reliability of the motion vector and the motion vector detection based on the result are repeated a plurality of times to obtain the final motion vector, the processing amount for the motion vector detection is significantly increased. To increase. In addition, when motion compensation is performed using the motion vector obtained in this way, the ratio of the prediction error signal power locally increases in the part where the block size is enlarged, so that the coding efficiency is improved especially in the flat part. Will fall. It should be noted that even when all the blocks are enlarged, the coding efficiency is reduced for the same reason.

Further, in the case where the input image is an interlaced image obtained by alternately performing odd field scanning indicated by a solid line and even field scanning indicated by a broken line as shown in FIG. When a motion vector is detected between odd and even fields that are out of phase, a motion vector that is completely unrelated to the motion of the subject is often detected due to the effects of aliasing distortion and phase shift, especially in blocks that include contours. .. In the interlaced image, FIG.
This is because there is no corresponding pixel in the adjacent field that is referred to when detecting the motion vector, as shown in FIG. The circles and the crosses in FIG. 7B indicate pixels obtained by scanning the solid line and the broken line in FIG. 7A, respectively.

When performing motion-compensated predictive coding with a fractional pixel accuracy such as 1/2 pixel accuracy or motion-compensated predictive coding incorporating an in-loop filter, a motion vector that is completely unrelated to the motion of the subject as described above is used. It is better to use a motion vector that reflects the motion of the subject, even if the accuracy is worse, than to use, the interpolated values represented by symbols ×, △, □ as shown in FIG. On the other hand, when the motion compensation is performed, the effect of the aliasing distortion or the deformation of the subject is reduced by the effect of the filter, so that the coding efficiency is often improved. In addition, ○ is the pixel value actually obtained,
X is the average value of two pixels A and B or C and D adjacent in the horizontal direction, Δ is the average value of two pixels A and C or B and D adjacent in the vertical direction, and □ is the average value of A, B, C and D. It is an average value.

However, conventionally, when such motion-compensated predictive coding with decimal pixel precision or motion-compensated predictive coding incorporating an in-loop filter is performed, there is no method for detecting an appropriate motion vector and high coding is performed. The efficiency was not obtained.

[0012]

As described above, in the conventional motion vector detection method in which the block size expansion and the motion vector detection are repeated to finally obtain a highly reliable motion vector, the motion vector detection is performed by the iterative calculation. There is a problem in that the coding efficiency in the flat part decreases due to a local increase in the ratio of the prediction error signal power when motion compensation is performed using motion vectors. there were.

Further, in the conventional motion vector detection method, when performing motion compensation predictive coding with decimal pixel precision or motion compensated predictive coding incorporating an in-loop filter, especially a motion vector suitable for an interlaced image is used. Is difficult to detect, and there is a problem that high coding efficiency cannot be obtained.

Therefore, a first object of the present invention is to detect a highly reliable motion vector between fields whose phases are shifted in the vertical direction without causing a large increase in the amount of processing due to repetitive calculations. It is to provide a motion vector detecting device capable of performing the above.

A second object of the present invention is to enable the detection of a motion vector which can improve the coding efficiency of the motion compensation predictive coding with decimal pixel precision and the motion compensation predictive coding incorporating an in-loop filter. To provide a motion vector detecting device.

[0016]

In order to achieve the first object of the present invention, the attribute of each block of an input image signal is determined by at least the proportion of the contour portion in the block, and the motion vector is determined based on the determination result. The feature is that the block size for detection is determined so as to increase as the proportion of the contour portion increases.

Further, in order to achieve the second object of the present invention, a new input field for each of the filtered signal and the unfiltered signal of the field image signal stored in the field memory. Detecting motion vectors of image signals (these are referred to as a first motion vector and a second motion vector, respectively), evaluating the first motion vector and the second motion vector, and outputting a motion vector having a better evaluation result. Is characterized by.

[0018]

Among the blocks of the input image signal, a block containing a large number of contours tends to adversely affect the coding efficiency in the motion compensation predictive coding when the motion vector detection is erroneous. According to the present invention, for such a block, the block size is enlarged at the time of motion vector detection, so that the reliability of motion vector detection is improved.

In this case, the expansion of the block size is not carried out sequentially according to the evaluation result of the reliability of the motion vector as in the prior art, but is uniquely determined from the attribute of each block. A large increase in volume is avoided.

Further, according to the present invention, when the motion vector is detected for each block of the input field image, in addition to the stored motion vector for the reference field image, the signal of the reference field image is filtered. The motion vector for the signal with reduced resolution is also detected, and the two are compared and the motion vector with the better evaluation result is used as the motion vector of the block.

As a result, when a motion vector is detected between fields whose phases are vertically shifted by the phase compensation and noise reduction by the filter, the motion compensation predictive coding with the decimal pixel precision and the in-loop filter are incorporated. The coding efficiency in the motion compensation predictive coding is improved. That is, when performing motion compensation with decimal pixel precision or motion compensation incorporating an in-loop filter, it is better to use a motion vector that reflects correct motion even if the evaluation result is worse than using a motion vector that is erroneously detected. Often, the prediction error signal power for the interpolated value or the value filtered by the in-loop filter is lowered, and the coding efficiency is often increased.

[0022]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment according to claim 1 of the present invention. The motion vector detection device of this embodiment includes an attribute determination circuit 11, a block size determination circuit 12,
It is composed of a frame memory or a field memory (hereinafter, simply referred to as a memory) 13 and a motion vector detection circuit 14. The image signal input to the terminal 10 is supplied to the attribute determination circuit 11, the memory 13, and the motion vector detection circuit 14.

In the attribute determination circuit 11, the attribute in the block divided into a certain size is determined from the input image signal, for example, by the ratio of the flat portion and the contour portion, and the result is determined by the block size determination circuit. Supply to 12.

FIG. 2 is a specific example of the attribute determination circuit 11, which is constructed by sequentially arranging a contour detection filter 21 using a Laplacian operator or the like, an absolute value circuit 22, a threshold circuit 23 and a counter 24. For example, a Laplacian value of the input image signal is obtained by the contour detection circuit 21, the absolute value of the Laplacian value is obtained by the absolute value circuit 22, and then the image signal is input to the threshold circuit 23. In the threshold circuit 23, a pixel whose absolute value is larger than a certain threshold is determined to be a contour portion. The number of pixels corresponding to the contour portion is counted for each block by the counter 24, and each block is classified into several groups according to the number. That is,
The output value of the counter 24 represents the ratio of the contour portion for each block.

As another configuration method of the attribute determination circuit 11,
For example, it is possible to use one that determines the ratio of the contour portion based on the activity in the block (the absolute value of the difference between each pixel value and the average value in the block or the sum of squares). In addition,
It is also possible to find the proportion of the flat portion in the block and convert the proportion of the contour portion from this.

In the block size decision circuit 12,
According to the attribute information determined by the attribute determination circuit 11, the enlargement ratio of the block size when the motion vector detection circuit 14 detects the motion vector, for example, how many pixels the block is enlarged vertically and horizontally as shown in FIG. .. Specifically, the block size is increased as the number of contoured blocks increases.

In FIG. 3, a block indicated by a solid line is a block for which motion compensation prediction is performed (referred to as a relevant block), and a block indicated by a broken line is a block for detecting a motion vector in a state where the block size is enlarged. Here, all detected motion vectors are motion vectors for the block in FIG.

The block size information determined by the block size determining circuit 12 in this way is supplied to the motion vector detecting circuit 14 as control information. On the other hand, the memory 13
The image signal stored in is input to the motion vector detection circuit 14 as a reference image signal. In the motion vector detection circuit 14, the input image signal input to the terminal 10 and the image signal stored in the memory 13 are divided into blocks according to the block size information from the block size determination circuit 12, and the motion vector is calculated by the block matching method. To detect. The information on the motion vector thus detected is output via the terminal 15.

In FIG. 4, P is a frame (field) in which motion compensation predictive coding is performed using a past frame (field) in the reference image, and B is a future frame (field) in the reference image as well as the past. Is a frame (field) for which motion compensation predictive coding is also performed, and I is a frame (field) that is coded within the frame (field). These are hereinafter referred to as P picture, B picture, and I picture. Further, the even field and the odd field in FIG.
It corresponds to the solid and dashed fields in (a).

Here, when detecting motion vectors from both directions as shown in FIG. 4A, the memory 13 may be prepared for a plurality of frames or a plurality of fields. The image signal stored in the memory 13 may be a locally decoded signal used in motion compensation predictive coding, that is, a signal obtained by decoding the coded signal.

Next, an embodiment according to claim 2 will be described with reference to FIG. The motion vector detection device of this embodiment includes a field memory 31, a filter 32, and a switch circuit 3.
3, a motion vector detection circuit 34 and an evaluation circuit 35.

The image signal (field image signal) input to the terminal 30 on a field-by-field basis is the field memory 3
1 and the motion vector detection circuit 24. The field memory 31 stores the input field image signal. The field image signal stored in the field memory 31 is supplied to the filter 32 and the switch circuit 33. In this case, when detecting motion vectors for a plurality of fields as shown in FIG. 4B, the field memories 31 may be prepared for a plurality of fields. The image signal stored in the field memory is
It may be a locally decoded signal used in motion compensation predictive coding.

In the filter 32, a signal between lines of the field image signal supplied from the field memory 31 is interpolated by an average value of upper and lower two pixels and the interpolated signal is supplied to the switch circuit 33. Switch circuit 33
Are switched in response to a request from the evaluation circuit 35, sequentially select the field image signals supplied from the field memory 31 and the filter 32, and supply the motion vector to the detection circuit 34.

In the motion vector detection circuit 34, the field image signals input via the switch circuit 33 are used as reference image (reference field image) signals, and these reference field image signals and a new input field image from the terminal 30. The motion vector is obtained by block matching with the signal. That is, a motion vector (first motion vector) indicating the motion of a new input field image signal with respect to the field image signal that has passed through the filter 32.
And a motion vector (second motion vector) indicating the motion of a new input field image signal with respect to the field image signal from the field memory 31 that does not pass through the filter 32 is detected.

The evaluation circuit 35 evaluates the first and second motion vectors, and outputs from the motion vector detection circuit 34 to the terminal 36 so that the one having the better evaluation result is the final motion vector of the block. Perform a tight control.
As a method of evaluating the motion vector in the evaluation circuit 35, for example, a method of outputting a vector having a small absolute sum or square sum in the block can be used.

However, if such a process is performed on all the reference field image signals, the calculation amount of motion vector detection is doubled, which is against the intended purpose of the present invention. Therefore, in this embodiment, the case where the above processing is performed is limited as follows.

Assuming the inter-field prediction as shown in FIG. 4B, a B picture needs a motion vector detecting ability for four reference field images, but an odd field P picture has a maximum of three references. Only motion vector detection for field images is performed. The effect of this embodiment is particularly great when detecting a motion vector between adjacent fields. Therefore, the filter 32
The block matching with the field that has passed through is limited to the case where the motion vector for the P picture of the odd field is detected using the P (I) picture of the even field as a reference image. By doing so, part of the ability to detect the four motion vectors prepared for the B picture when searching for an odd field of the P picture can be diverted, and the hardware scale does not increase. ..

[0038]

According to the present invention, when a motion vector for a moving image signal is detected, a highly accurate motion can be achieved between vertically phase-shifted fields without a large increase in the amount of processing due to repeated calculations. Vectors can be detected.

Further, according to the present invention, a highly reliable motion vector which does not adversely affect the coding efficiency of the motion-compensated predictive coding with the decimal pixel precision and the motion-compensated predictive coding incorporating the in-loop filter. It becomes possible to detect.

[Brief description of drawings]

FIG. 1 is a block diagram of a motion vector detection device according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a specific configuration example of an attribute determination circuit in FIG.

FIG. 3 is a diagram showing enlargement of blocks in the embodiment.

FIG. 4 is a diagram showing a relationship between each frame image or each field image used for motion compensation predictive coding.

FIG. 5 is a block diagram of a motion vector detecting device according to another embodiment of the present invention.

FIG. 6 is a diagram for explaining the principle of a motion vector detection method by block matching.

FIG. 7 is a diagram showing a relationship between fields in an interlaced image.

FIG. 8 is a diagram for explaining motion compensation with decimal pixel precision.

[Explanation of symbols]

10 ... Image signal input terminal 11 ... Attribute determination circuit 12 ... Block size determination circuit 13 ... Frame (field) memory 14 ... Motion vector detection circuit 15 ... Motion vector information output terminal 21 ... Contour detection filter 22 ... Absolute value circuit 23 ... Threshold Circuit 24 ... Counter 30 ... Image signal input terminal 31 ... Field memory 32 ... Filter 33 ... Switch circuit 34 ... Motion vector detection circuit 35 ... Evaluation circuit 36 ... Motion vector information output terminal

Claims (2)

[Claims] 1. A storage means for storing a predetermined amount of an input image signal, and a motion vector detection means for detecting a motion vector indicating a movement of a new input image signal with respect to the image signal stored in the storage means, in units of blocks of a predetermined size. Determination means for determining the attribute of each block of the input image signal using at least the proportion of the contour portion, and a block size for motion vector detection in the motion vector detection means based on the determination result of this determination means. And a block size determining means for determining so as to increase as the proportion of the contour portion increases. 2. Storage means for storing at least one field of an input field image signal, filtering means for filtering the field image signal stored in this storage means, and filtering processing by this filtering means. A first motion vector indicating the motion of the new input field image signal with respect to the field image signal, and a second motion vector indicating the motion of the new input field image signal with respect to the field image signal stored in the storage means. Motion vector detecting means for evaluating the first and second motion vectors detected by this means, and evaluating means for outputting the motion vector having the better evaluation result, the motion of the image. Vector detection device.