JP2006217449A - Motion vector detecting method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve encoding efficiency by suppressing selection errors, when one is selected from either the frame motion vector and field motion vector of which are to serve as candidates. <P>SOLUTION: A motion vector detecting method in a motion compensation predicting encoding detects candidates of the frame motion vector and the field motion vector for a macro block of an input present frame, and selects by vector selection means 18 any one of the vectors, by using a comparison result by evaluation value comparison means 17 evaluated values attached to each vector. At this time, weight factors, that change, in response to the total movement amount and reliability from total movement amount detection means 14, are added to the one estimation value in weight addition means 16, and are compared in the evaluation value comparison means 17. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a motion vector detection method used for motion compensated predictive coding of moving image signals, particularly interlaced image signals.

  In recent years, a motion-compensated prediction interframe coding method has been used as a method for efficiently compressing information of a moving image signal. This scheme includes predicting motion between image frames.

  The motion compensated prediction interframe coding method detects a pixel block of a past or future reference frame having the highest correlation with a macroblock to be coded (hereinafter referred to as a current macroblock), and detects the current macroblock. The difference between the obtained pixel blocks (hereinafter referred to as reference blocks) is encoded. When transmitting the encoded data, a motion vector indicating the amount of motion of the reference block from the current macroblock is encoded and transmitted along with the encoded difference data.

  In a general motion vector detection method, the current macroblock performs frame motion prediction and field motion prediction on a certain reference frame. Here, the current macroblock is a block made up of, for example, 16 × 16 pixels generated by dividing the image frame data into blocks.

  When performing frame motion prediction, the current macroblock performs a block matching check on a reference frame within a predetermined search range, and specifies a reference block having the highest correlation. At this time, the spatial displacement of the identified reference block from the current macroblock is referred to as a frame motion vector. The correlation index used when the frame motion vector is specified is set as the frame evaluation value. In the block matching inspection, the sum of absolute pixel differences between corresponding 16 × 16 pixel blocks is used as an evaluation value, and this evaluation value is obtained for all reference blocks in the search range, and the evaluation value is minimized. In this method, the reference block is determined to have the highest correlation.

  When performing field motion prediction, first, the current macroblock is divided into two field blocks of 16 × 8 pixels. Here, the two fields constituting the frame are called a top field and a bottom field, and the field block divided into two is called a current top field block and a current bottom field block.

  The current top field block performs a block matching check on the top field of the reference frame, identifies the reference block having the highest correlation, and determines the first top field vector and the first top field evaluation value. Similarly, the current top field block performs a block matching check on the bottom field of the reference frame to obtain a second top field vector and a second top field evaluation value.

  Similarly, the current bottom field block performs a block matching check on the top field of the reference frame to determine a first bottom field vector and a first bottom field evaluation value. Similarly, the current bottom field block performs a block matching check on the bottom field of the reference frame to determine a second bottom field vector and a second bottom field evaluation value.

  Thus, one frame vector and four field vectors for the reference frame are obtained as candidates for one current macroblock. When encoding finally, select one of these frame vectors or a field vector consisting of one top field vector and one bottom field vector, and if there are two reference frames, One vector is selected by selecting a vector for one reference frame or a bi-directional reference vector composed of both reference frames and used for encoding.

  The determination method used at this time is generally such that a vector having the smallest evaluation value is selected from the evaluation values of all candidate vectors.

  Now, an implementation method of the motion vector detection method described above will be described by taking test model 5 (TM5) of MPEG2 (Moving Picture Coding Experts Group Phase 2), which is generally known as a technique, as an example.

  FIG. 7 is a block diagram showing the configuration of an apparatus using this conventional motion vector detection method.

  The frame motion vector detection means 102 is a block for a frame macroblock having a current frame input from the input terminal 100 and all frame reference blocks within a set vector search range having a reference frame input from the input terminal 101. A matching check is performed, a block having the smallest evaluation value Df is selected, and a motion vector MVf indicating the block position is detected.

  Next, the field motion vector detecting means 103 receives all the top fields in the set vector search range including the field macroblock having the current frame input from the input terminal 100 and the reference frame input from the input terminal 101. Block matching inspection is performed on the reference block and the bottom field reference block, and the four types of field motion vectors and their evaluation values are determined as described above. Then, the optimum set of the top field vector MVt and the bottom field vector MVb is selected, and the evaluation value Di is output. The evaluation value Di is obtained as the sum of the evaluation value of the selected top field vector MVt and the evaluation value of the bottom field vector MVb.

  The evaluation value is the sum of absolute pixel differences between the current macroblock and the reference block.

  The evaluation value comparison unit 104 compares the frame motion vector evaluation value Df determined above and the field motion vector evaluation value Di, and determines that the motion vector having the smaller evaluation value is optimal. A control signal for selecting the motion vector is output to the vector selection means 105.

  The vector selection unit 105 receives the determined frame vector MVf and the set of the field vectors MVt and MVb, and selects one motion vector according to the control signal from the evaluation value comparison unit 104. The selected motion vector is output to the output terminal 106. The selected motion vector is finally used for motion compensation prediction encoding.

Further, in the motion vector detection device disclosed in Patent Document 1, a method for detecting a vector candidate having the smallest generated code amount from among vector candidates that take an evaluation value near the minimum value even if it is not the minimum value is shown. Yes. The main purpose of this method is to prevent an increase in the amount of code when encoding and transmitting a vector due to variations in the vector resulting from the reliability of the evaluation value not being very high. Perform motion detection within the search range, calculate the minimum evaluation value, then set a range slightly larger than the minimum value, select a candidate vector that takes an evaluation value within the range from all candidate vectors, If there is a candidate vector that matches the common vector obtained in the macroblock, it is adopted.
JP 2000-152243 A

  However, the conventional motion vector detection method as indicated by TM5 has the following problems.

  In this conventional motion vector detection method, for example, the difference between the frame evaluation value and the field evaluation value is remarkably generated in a clear part such as an edge part of a moving object. in the region, I do not know the difference between the frame evaluation value with the field evaluation value either because the hard output is selected. In particular, when the input image is a still image or when the entire screen is moving slowly, the frame motion can be detected because there is little deformation between the reference frame and the current frame, and matching between pixels in the frame block is easy. It is easy to become. However, there are many cases where a field vector is selected in a uniform region having no features as described above. In such a case, when a field vector is employed, (1) the vector code amount must be transmitted more than the frame vector. (2) When viewed in a partial area, there is a problem that non-uniform distortion occurs due to frequent changes in the reference method even though the images are similar between consecutive frames. As described above, the evaluation value comparison itself does not include information such as image characteristics, and the problem is that an optimal motion vector cannot be obtained in that the comparison is always equivalent.

  On the other hand, in the case of an image in which the entire screen is moving greatly, there is a time difference between the fields, so that a picture shift occurs between the top field and the bottom field constituting the frame block. As long as this is not equal to such a deviation of the reference frame, the probability that the frame vector can be searched decreases. However, a frame vector may be selected in the uniform image area.

  In addition, as described above, the motion vector detection device disclosed in Patent Document 1 detects a vector code having the smallest generated code amount among vector candidates that take an evaluation value near the minimum value even if it is not the minimum value. However, the purpose is to improve the image quality by reducing the amount of code according to the motion vector coding algorithm, and it is characterized by matching the vector type and vector length to the immediately preceding macroblock as much as possible. It does not take into account the characteristics of the entire image or the distortion non-uniformity between successive frames.

  Therefore, an object of the present invention is to improve the image quality by making it possible to select an appropriate motion vector in the generated code amount / distortion uniformization even in a portion having no pattern feature based on the motion amount of the entire image.

  It is another object of the present invention to perform variable control for each divided area of an image and to select an appropriate motion vector in the generated code amount / distortion uniformization according to a stationary part / motion part in one frame so as to improve image quality. .

  The present invention takes the following means in order to solve the above problems.

  As a first solution, a motion vector detection method according to the present invention is a method of dividing an input image frame into predetermined blocks and detecting a motion vector representing motion from another reference frame for each block, A frame motion vector for the reference frame for the block, a field motion vector for the first or second field of the reference frame for each field block obtained by separating the block into two field blocks, and a frame evaluation value associated with the frame motion vector; The field evaluation value associated with the field motion vector is calculated and compared. When one of the motion vectors is selected, the frame evaluation value and the field evaluation value are changed according to the magnitude of the overall movement amount of the image. Characterized by comparing to at least one of To.

  In the above, the total movement amount is the displacement point at which the sum of absolute values of the differences between the matching pixels is minimized by displacing the entire image between the current image to be encoded and the reference image used for motion compensation prediction. As determined. In addition, the current frame and the reference frame used when detecting the total movement amount may be reduced images that are subsampled at a fixed thinning rate.

  In the above, the weight assigned to the evaluation value also changes depending on the reliability with respect to the total movement amount. In particular, the frame motion vector changes such that the smaller the overall movement amount and the higher the reliability, the more preferentially the frame motion vector is selected.

  In the above, the reliability is calculated based on a sum of absolute differences, a sum of squares of differences, or a correlation coefficient between matching pixels of the current frame and the reference frame at the displacement point indicated by the total movement amount. .

  In the above, the evaluation value comparison is to select a frame motion vector when the condition of Fa ≦ Fb + K is satisfied, where Fa is a frame evaluation value, Fb is a field evaluation value, and K is a weight. When the reliability is judged to be high or low by comparing the reliability with a predetermined threshold, and when the reliability is judged to be low, the weight K is limited to the minimum value, and the reliability is judged to be high In addition, the weight K is set to a larger value as the total movement amount is smaller and compared. The comparison of evaluation values selects a frame motion vector when a condition of Fa ≦ Fb × M is satisfied, where Fa is a frame evaluation value, Fb is a field evaluation value, and M is a weight. By comparing the degree with a predetermined threshold value, it is determined whether the reliability is high or low. When it is determined that the reliability is low, the weight M is limited to the minimum value, and when the reliability is determined to be high For comparison, the weight M may be set to a larger value as the total movement amount is smaller.

  As a second solution, a motion vector detection method according to the present invention is a method of dividing an input image frame into predetermined blocks and detecting a motion vector representing a motion from another reference frame for each block, A frame motion vector for the reference frame for the block, a field motion vector for the first or second field of the reference frame for each field block obtained by separating the block into two field blocks, and a frame evaluation value associated with the frame motion vector; The field evaluation value associated with the field motion vector is calculated and compared, and when one motion vector is selected, the image is divided into a plurality of regions, and the amount of movement is obtained for each divided region, The weight varies according to the amount of movement of the area that contains And comparing subjected to at least one of the value and the field evaluation value.

  In the above, the movement amount is determined as a displacement point at which the sum of absolute differences between pixels that are displaced and matched between the image of the divided region of the current frame and the reference frame used for motion compensation prediction is minimum. The Further, the current frame and the reference frame used when detecting the movement amount may be reduced current images that are subsampled at a fixed thinning rate.

  In the above, the weight given to the evaluation value also changes depending on the reliability with respect to the movement amount of the area including the block. In particular, the frame motion compensation prediction method is changed so as to be preferentially selected as the movement amount is small and the reliability is large.

  In the above, the reliability is calculated based on the sum of absolute difference values, sum of squared differences, or correlation coefficient between the matching pixels between the image of each area of the current frame and the reference frame at the displacement point indicated by the movement amount of each area. It is characterized by being.

  In the above, the evaluation value comparison is to select a frame motion vector when the condition of Fa ≦ Fb + K is satisfied, where Fa is a frame evaluation value, Fb is a field evaluation value, and K is a weight. When the reliability is judged to be high or low by comparing the reliability with a predetermined threshold, and when the reliability is judged to be low, the weight K is limited to the minimum value, and the reliability is judged to be high The weight K is set to a larger value as the total movement amount is smaller, and the comparison is performed. The comparison of evaluation values selects a frame motion vector when a condition of Fa ≦ Fb × M is satisfied, where Fa is a frame evaluation value, Fb is a field evaluation value, and M is a weight. By comparing the degree with a predetermined threshold value, it is determined whether the reliability is high or low. When it is determined that the reliability is low, the weight M is limited to the minimum value, and when the reliability is determined to be high For comparison, the weight M may be set to a larger value as the total movement amount is smaller.

  The motion vector detection method of the present invention is capable of reducing detection errors even in a portion having no pattern feature based on the amount of motion of the entire image, and selecting an appropriate motion vector. And uniform image quality between frames.

  In addition, using reliability, it is possible to give a large weight when the detection result of the total movement amount of the image can be trusted, and even when the total movement amount cannot be detected correctly, a proper motion vector can be selected without malfunction. it can.

  In addition, by detecting the amount of movement and the reliability for each divided region of the image, it is possible to select an appropriate motion vector according to the image characteristics in the screen, and the encoding efficiency can be improved by finer control and Uniform image quality between frames can be realized.

  Embodiments of a motion vector detection method according to the present invention will be described below in detail with reference to the drawings.

(Embodiment 1)
FIG. 1 is a block diagram showing a configuration of an apparatus to which a motion vector detection method according to Embodiment 1 of the present invention is applied. In FIG. 1, 10 is a current frame input terminal, 11 is a reference frame input terminal, 12 is a frame motion vector detection means, 13 is a field motion vector detection means, 14 is an overall movement amount detection means, 15 is a weight coefficient generation means, 16 Is a weight addition means, 17 is an evaluation value comparison means, 18 is a vector selection means, and 19 is a motion vector output terminal.

  The operation of the motion vector detection apparatus configured as described above will be described below.

  First, the frame motion vector detection means 12 includes a certain frame macroblock in the current frame input from the current frame input terminal 10 and all within a set vector search range in the reference frame input from the reference frame input terminal 11. A block matching check is performed on the frame reference block, a block having the smallest evaluation value Df is selected, and a motion vector MVf indicating the block position is detected.

  Similarly, the field motion vector detection means 13 is input from the reference frame input terminal 11 and the top field macroblock and the bottom field macroblock obtained by separating the macroblock of the current frame input from the current frame input terminal 10 by field. A block matching check is performed on all top field reference blocks and bottom field reference blocks within a certain set vector search range in the reference frame, and a field motion vector is detected. There are various possible methods for determining the field motion vector. For example, the top field motion vector MVt indicating the position of the block having the minimum evaluation value Dt among the field reference blocks referred to by the top field macroblock, and the bottom field macro A combination with the bottom field motion vector MVb indicating the position of the block having the minimum evaluation value Db among the field reference blocks referred to by the block is selected and set as the field motion vector MVi. Further, the sum of Dt and Db is obtained as the evaluation value of the field motion vector, and this is set as the field motion vector evaluation value Di. The evaluation value is, for example, the total sum of pixel difference absolute values between the current macroblock and the reference block.

  The frame motion vector MVf and the field motion vector MVi determined in this way are input to the vector selection means 18. On the other hand, the frame motion vector evaluation value Df is input to the evaluation value comparison means 17, and the field motion vector evaluation value Di is input to the weight addition means 16. The weight addition unit 16 adds the weight coefficient generated by the weight coefficient generation unit 15 to the field evaluation value Di, and outputs the result Di ′ to the evaluation value comparison unit 17. The evaluation value comparison means 17 compares the input frame motion vector evaluation value Df with the field motion vector evaluation value Di ′ to which the weighting coefficient is added, and selects a control signal for selecting a motion vector type that takes a smaller value. It outputs to the vector selection means 18. The vector selection means 18 selects a motion vector indicated by the control signal and outputs it to the motion vector output terminal 19.

  The total movement amount detection means 14 detects the movement amount of the entire image from the current frame input from the current frame input terminal 10 and the reference frame input from the reference frame input terminal 11. Then, the detected total movement amount GV is output. Further, when calculating the reliability coefficient S representing the reliability of the total movement amount GV, the reliability coefficient S is also output. The weight coefficient generation means 15 generates a predetermined weight coefficient W according to the total movement amount GV and outputs it to the weight addition means 16. When considering the reliability, the weight coefficient generation means 15 further inputs the reliability coefficient S, generates a predetermined weight coefficient W according to the reliability coefficient S, and outputs it to the weight addition means 16.

  FIG. 2 is a detailed block diagram of the total movement amount detection means 14.

  In FIG. 2, 50 and 51 are frame memories, 52 is an address control means, 53 is a matching inspection means, 54 is a movement amount determination means, 55 is a reliability calculation means, 56 is an overall movement amount output terminal, and 57 is a reliability coefficient. Output terminal.

  The current frame input from the current frame input terminal 10 and the reference frame input from the reference frame input terminal 11 are input to the frame memories 50 and 51, respectively. The matching checking means 53 performs a pattern matching check on the entire frame of the current frame data obtained from the frame memory 50 and the reference frame data obtained from the frame memory 51. At this time, the address control means 52 controls reading of data from each frame memory so that pattern matching can be performed by shifting the reference frame from the current frame by a predetermined amount of movement. Further, the amount of shift may be shifted by a predetermined amount vertically and horizontally. Particularly, the object of the present invention is to detect an image feature having a small overall movement amount. It is only necessary to inspect the minimum deviation in relation to

  The matching checking means 53 calculates the sum of absolute pixel difference values for the entire frame for each shift amount. The movement amount determination unit 54 sets the movement amount indicating the shift position that takes the minimum value from the total sum of pixel differences obtained from all the shift amounts input from the matching inspection unit 53 as the total movement amount GV to the total movement amount output terminal 56. Output. On the other hand, the total movement amount GV is input to the reliability calculation means 55. The reliability calculation means 55 receives the reference frame shifted by the shift amount referred to by the total movement amount GV, and obtains the reliability with the current frame input at the same time. As the reliability index, a pixel difference absolute value sum, a correlation coefficient, a pixel difference square sum, or the like between matching pixels may be used. In either case, the reliability is converted into a reliability coefficient S that takes a larger value as the reliability is higher.

  In the case where the reliability coefficient S takes a smaller value as the reliability is higher, the same effect can be obtained if the present description is read so as to be adapted thereto.

  The index may be output as it is, and may be configured to correspond to the reliability indicated by the index when it is used. When the pixel difference absolute value sum is used as the reliability index, the value used when determining the total movement amount GV may be output as it is, and in this case, the reliability calculation means 55 may be unnecessary.

  FIG. 3 is a conceptual diagram for explaining a specific operation of the entire movement amount detection.

  The solid line rectangle in FIG. 3 is the current frame, and the broken line rectangle is the reference frame. An image in which the reference frame is shifted in each direction based on the upper left corner of the current frame is drawn. A rectangular portion hatched for each shift amount is a pixel subjected to pattern matching inspection. Although only nine patterns are drawn, this shows the maximum shift amount in each direction, and there is also an intermediate shift amount between them. The shift amount may be arbitrarily determined as described above. It is necessary to pay attention to the fact that the evaluation point (number of pixels) of the sum of absolute differences of pattern matching results differs for each shift amount, and it is useful to perform normalization based on the number of matching pixels.

  Next, operations of weight addition and vector selection by the weight coefficient generation means 15, weight addition means 16 and evaluation value comparison means 17 will be described.

  As described above, the total movement amount GV and the reliability coefficient S have already been determined and input to the weight coefficient generation means 15. As an example, the reliability coefficient S will be described assuming that the correlation coefficient calculation result is used as it is as the reliability coefficient.

  First, the evaluation value comparison means 17 generates a control signal so as to select a frame motion vector when the following conditions are satisfied.

(Formula 1-1) Df ≦ Di + W
Further, when the reliability is used, a control signal is generated so as to select a frame motion vector when the following condition is further satisfied.

(Formula 1-2) S ≧ A (-1 ≦ A ≦ 1 real number)
When the characteristics are set in this way, the weighting factor W is obtained as follows.

(Equation 1-3) W = B−a × GV (B ≧ 0 real number, a is an arbitrary real number, W ≧ 0)
Further, the method of weight addition and vector selection may be configured as follows.

  First, the evaluation value comparison means 17 generates a control signal so as to select a frame motion vector when the following conditions are satisfied.

(Formula 2-1) Df ≦ Di × W
Further, when the reliability is used, a control signal is generated so as to select a frame motion vector when the following condition is further satisfied.

(Formula 2-2) S ≧ A (-1 ≦ A ≦ 1 real number)
When the characteristics are set in this way, the weighting factor W is obtained as follows.

(Formula 2-3) W = B−a × GV (B ≧ 0 real number, a is an arbitrary real number, W ≧ 1)
Further, the conditions of the above (Expression 1-2) and (Expression 2-2) using the reliability S may be functions. For example, the weight W is determined by a function as shown in FIG.

  FIG. 4 is a diagram showing calculation elements of the weighting factor W. Further, the comparison condition of the comparative evaluation value comparison means 17 is to calculate the weighting coefficient W that is suitable for the case of the above (Equation 1-1).

  FIG. 4A shows a function for calculating the weight element W1 based on the magnitude of GV. Here, the threshold value TH1 and the weight element maximum value P can be arbitrarily set. The function for determining the weight element W1 is set such that the smaller the GV, the larger the weight element W1. FIG. 4B shows a function for calculating the weight element W2 based on the magnitude of S. Here, the threshold value TH2 and the weight element maximum value Q can be arbitrarily set. The function for determining the weight element W2 is set so that the greater the S, the greater the weight element W2. The weight coefficient W is calculated by the product of the weight element W1 and the weight element W2.

  The calculation function of the weight element W1 and the weight element W2 can be variously modified.

  Although the weighting factor W is calculated by the product of the weighting element W1 and the weighting element W2, the weighting factor W may be calculated by the sum of the weighting element W1 and the weighting element W2. In this case, the weighting factor W What is necessary is just to select the calculation function of W1 and the weight element W2 to match this.

  By performing motion vector selection as described above, the frame motion vector is more easily selected when the movement amount of the entire image is small and its reliability is high, and the reliability is high when the movement amount of the entire image is large. If it is low, the rate at which the frame motion vector is selected can be lowered.

  Note that the current frame and the reference frame input to the total movement amount detection means 14 in the above description may use reduced images subsampled at a predetermined thinning rate. As a result, it is possible to reduce the amount of calculation spent on the total movement detection according to the thinning rate.

  When a plurality of reference frames are used and the time distance between the current frame and each reference frame is different, the threshold TH1 used when determining the weight element W1 based on the total movement amount GV is set as the time distance between the reference frames. You only need to set each value to fit.

  When a plurality of reference frames are used and a bidirectional reference motion vector can also be selected, the above-described weights may be added to the field bidirectional motion vector. In this case, the weighting factor may be determined using the total movement amount GV and the reliability S of each reference frame. For example, the total movement amount GV and reliability S of one reference frame may be used as they are.

  As described above, based on the amount of motion of the entire image, it is possible to adaptively reduce detection errors even in a portion without a pattern feature, and to select an appropriate motion vector. Improvement and uniform image quality between frames can be realized. In addition, since reliability can be used to apply a large weight when the detection result of the total movement amount of the image can be trusted, an appropriate motion vector is selected without malfunction even when the total movement amount cannot be detected correctly. be able to.

(Embodiment 2)
Next, the operation of the apparatus to which the motion vector detection method according to the second embodiment of the present invention is applied will be described. FIG. 5 is a block diagram showing a configuration of an apparatus to which the motion vector detection method according to the second embodiment of the present invention is applied. In FIG. 5, 30 is a movement amount detection means, 31 is a weight coefficient generation means, and the other components have the same functions as those in FIG.

  The operation of the motion vector detection apparatus configured as described above will be described below.

  The frame motion vector MVf, the field motion vector MVi, and the evaluation values Df and Di are determined in the same manner as in the first embodiment. The image movement amount detection and weighting coefficient generation and comparison methods are different. Hereinafter, a different part from Embodiment 1 is demonstrated.

  The movement amount detection means 30 detects the movement amount GVn for each of a plurality of regions of the current frame using the current frame input from the current frame input terminal 10 and the reference frame input from the reference frame input terminal 11. The movement amount detecting means 30 includes means for dividing the input current frame into a plurality of areas. As described with reference to FIG. 2 in the first embodiment, the reference frame is shifted from the current frame by a predetermined amount of movement, and a pattern matching check with the current frame is performed. In this case, the pattern matching check is performed for each divided region of the current frame. Then, the movement amounts GV1 to GVn (n is the number of divided areas) obtained in each area are output.

  FIG. 6 is a conceptual diagram for explaining a method for detecting the movement amounts of the plurality of regions. FIG. 6 shows a state in which the pattern matching inspection is performed by shifting the reference frame indicated by the broken rectangle by a certain amount of movement with respect to the current frame indicated by the solid rectangle. Such pattern matching inspection is performed for all the inspection ranges set in advance, and movement amounts GV1 to GVn at which the sum of pixel difference absolute values becomes the minimum value are calculated for each region. Further, when calculating reliability coefficients corresponding to the movement amounts GV1 to GVn for each region, the reliability coefficients S1 to Sn are also output.

  The weighting factor generating unit 31 generates a predetermined weighting factor W according to the movement amount GVr corresponding to the region number r including the macroblock for which a motion vector is currently detected, and outputs it to the weighting unit 16. When considering the reliability, the weight coefficient generation means 31 further inputs a reliability coefficient Sr, generates a predetermined weight coefficient W according to the reliability coefficient Sr, and outputs it to the weight addition means 16. The weighting factor generating means 31 is different from the operation of the first embodiment in that the weighting factor W is generated based on the movement amount and reliability according to the region to which the macroblock belongs. However, the process of determining the weighting coefficient from the movement amount and the reliability for each region is the same as in the first embodiment.

  The weight addition unit 16, the evaluation value comparison unit 17, and the vector selection unit 18 perform the same operation as in the first embodiment, select a motion vector, and output it to the output terminal 19.

  As described above, based on the amount of movement for each area of the image, even in the case of an image having a stationary part / motion part in a frame, the detection error is adaptively reduced in a part having no pattern feature, and an appropriate motion can be reduced. The vector can be selected, and the encoding efficiency can be improved and the image quality can be made uniform between frames. In addition, using reliability, it is possible to apply a large weight when the detection result of the total movement amount of the image can be trusted, so even when the movement amount cannot be detected correctly, an appropriate motion vector can be selected without malfunction. Can do.

  As described above, according to the present invention, motion vector detection errors can be reduced in accordance with image characteristics, and effects such as reduction of the amount of generated codes and equalization of distortion between consecutive frames are brought about. Therefore, an apparatus such as a high-quality video device using motion compensation predictive coding using motion vector detection can be realized, which is very useful industrially.

1 is a block diagram showing a configuration of an apparatus to which a motion vector detection method according to Embodiment 1 of the present invention is applied. The block diagram which shows the detailed structure of the whole movement amount detection means in the same motion vector detection method Conceptual diagram for explaining the operation of detecting the total movement amount in the motion vector detection method Conceptual diagram illustrating an example of a weight coefficient generation method in the motion vector detection method A block diagram showing a configuration of an apparatus to which a motion vector detection method according to a second embodiment of the present invention is applied. Conceptual diagram explaining operations of area division and movement amount detection in the same motion vector detection method Block diagram showing the configuration of an apparatus to which a conventional motion vector detection method is applied

Explanation of symbols

10, 100 Current frame input terminal 11, 101 Reference frame input terminal 12, 102 Frame motion vector detection means 13, 103 Field motion vector detection means 14 Overall movement amount detection means 15, 31 Weight coefficient generation means 16 Weight addition means 17, 104 Evaluation value comparison means 18, 105 Vector selection means 19, 106 Motion vector output terminal 30 Movement amount detection means

Claims (16)

A method of dividing an input image frame into predetermined blocks and detecting a motion vector representing a motion from another reference frame for each of the blocks, obtaining a frame motion vector for the reference frame for the block, For each field block, a field motion vector for the first or second field of the reference frame is obtained, a frame evaluation value associated with the frame motion vector, and a field evaluation associated with the field motion vector When calculating one of the motion vectors and selecting one of the motion vectors, a weight that changes in accordance with the amount of overall movement of the image is attached to at least one of the frame evaluation value and the field evaluation value. Motion vector detection characterized by comparison Law. The total movement amount is a displacement point at which the total sum of absolute values of differences between pixels to which the entire image is displaced and matched between the current frame to be encoded and a reference frame used for motion compensation prediction is minimized. The motion vector detection method according to claim 1, wherein: 3. The motion vector detection method according to claim 2, wherein the current frame and the reference frame used when detecting the total movement amount are reduced images subsampled at a fixed thinning rate. 2. The motion vector detection method according to claim 1, wherein the weight further changes depending on a reliability with respect to the total movement amount. 5. The motion vector detection method according to claim 4, wherein the weight changes such that a frame motion vector is preferentially selected as the total movement amount is small and the reliability is large. The reliability is calculated based on a sum of absolute differences, a sum of squared differences, or a correlation coefficient between matching pixels between the current frame and the reference frame at a displacement point indicated by the total movement amount. The motion vector detection method according to claim 4. The evaluation value comparison is to select a frame motion vector when the condition of Fa ≦ Fb + K is satisfied, where Fa is the frame evaluation value, Fb is the field evaluation value, and K is the weight. By comparing the reliability with a predetermined threshold value, it is determined whether the reliability is high or low. When it is determined that the reliability is low, the weight K is limited to the minimum value, and it is determined that the reliability is high. 5. The motion vector detection method according to claim 4, wherein the weight K is set to a larger value as the total movement amount is smaller. The comparison of the evaluation values is to select a frame motion vector when the condition of Fa ≦ Fb × M is satisfied, where Fa is the frame evaluation value, Fb is the field evaluation value, and M is the weight. If the reliability is high or low by comparing the reliability with a predetermined threshold, and if it is determined that the reliability is low, the weight M is limited to a minimum value and the reliability is high. 5. The motion vector detection method according to claim 4, wherein, when judged, the weight M is set to a larger value as the total movement amount is smaller and compared. A method of dividing an input image frame into predetermined blocks and detecting a motion vector representing a motion from another reference frame for each of the blocks, obtaining a frame motion vector for the reference frame for the block, Are divided into two field blocks, a field motion vector for the first or second field of the reference frame is obtained for each field block, a frame evaluation value associated with the frame motion vector, and a field associated with the field motion vector When calculating and comparing the evaluation value and selecting one motion vector, the image is divided into a plurality of regions, the amount of movement is obtained for each divided region, and the movement of the region including the block is calculated. The weight that changes in accordance with the amount is set to the frame evaluation value and the file. The motion vector detection method and comparing subjected to at least one of the field evaluation value. The amount of movement is obtained as a displacement point at which the sum of absolute differences between the pixels that are displaced and matched between the image of the divided region of the current frame and the reference frame used for motion compensation prediction is minimum. The motion vector detection method according to claim 9. 11. The motion vector detection method according to claim 10, wherein the current frame and the reference frame used when detecting the movement amount are reduced current images subsampled at a predetermined thinning rate. The motion vector detection method according to claim 9, wherein the weight further changes in accordance with a reliability with respect to the movement amount of an area including the block. 13. The motion vector detection method according to claim 12, wherein the weight changes so that a frame motion compensation prediction method is preferentially selected as the movement amount is small and the reliability is large. . The reliability is calculated based on a sum of absolute differences, a sum of squared differences, or a correlation coefficient between matching pixels between an image of a divided region of the current frame at the displacement point indicated by the movement amount and the reference frame. The motion vector detection method according to claim 12. The evaluation value comparison is to select a frame motion vector when the condition of Fa ≦ Fb + K is satisfied, where Fa is the frame evaluation value, Fb is the field evaluation value, and K is the weight. By comparing the reliability with a predetermined threshold value, it is determined whether the reliability is high or low. When it is determined that the reliability is low, the weight K is limited to the minimum value, and it is determined that the reliability is high. 13. The motion vector detection method according to claim 12, wherein the weight K is set to a larger value as the total movement amount is smaller. The comparison of the evaluation values is to select a frame motion vector when the condition of Fa ≦ Fb × M is satisfied, where Fa is the frame evaluation value, Fb is the field evaluation value, and M is the weight. If the reliability is high or low by comparing the reliability with a predetermined threshold, and if it is determined that the reliability is low, the weight M is limited to a minimum value and the reliability is high. 13. The motion vector detection method according to claim 12, wherein when the determination is made, the weight M is set to a larger value as the total movement amount is smaller and compared.

Images