JP2000050282A - Motion detector, motion detection method and recording medium with its program recorded therein - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain the difference between a motion detection object block obtained by sequentially segmenting a block from an image and a prediction block in a reference image through a simple hardware configuration in which same processing is repeated for each processing cycle in the unit of blocks. SOLUTION: An evaluated position closer to a range of a motion of an adjacent block and with a small difference is made easy to be left in a minimum value and evaluation position storage device 60 as a prediction block providing a minimum difference. First a corrected correction consolidated difference is generated, which is obtained by summing a difference A of a single detection object block (output signal of a difference evaluation device 10) and a difference Bf of an adjacent block, whose motion vector is the same as that of the detection object block and processing the sum. A difference A or the corrected consolidated difference, whichever is the smaller is used as the final difference. Then a minimum final difference and the evaluation position providing the minimum difference are left in the minimum value and evaluation position storage device 60 for each processing cycle in the unit of blocks, and they are outputted at the end of the processing cycle in the unit of blocks.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] The present invention relates to a method for producing a target block, comprising:
A motion detector that predicts from the image in front of or behind in time series the image that includes it, and calculates a vector from the predicted block (predicted block) to the target block as a motion vector; The present invention relates to a motion detection method and a recording medium on which a program for the method is recorded.

[0002] The motion detection referred to here is not only important as a core function of a video compressor using motion compensation, but also a useful function for image feature extraction and recognition.

[0003]

2. Description of the Related Art FIG. 1 is a diagram for explaining a conventional motion search. In the figure, reference numeral 1 denotes a reference image, 2 denotes a motion detection target image,
Reference numeral 3 denotes a search area, and reference numeral 4 denotes a motion detection target block.

Conventionally, in motion detection, as shown in FIG. 1, only a detection target block 4 is used as a difference evaluation template, and a block having the smallest difference is searched for in a fixed area (search area) of the reference image 1. Then, the result was predicted as the movement source block.

In this method, when the brightness and the pattern are uniform, such as in the sky, on a wall, or on a lawn, the source block does not necessarily have the minimum degree of difference.
In some cases, the source block cannot be predicted correctly, and the motion vector varies in spite of uniform motion.

In particular, when a motion search is performed on a sub-sampled image in order to reduce the amount of calculation, the search accuracy is reduced, and this tendency appears strongly. Nevertheless, when motion detection is used for motion compensation in coding, even if the source block is erroneously detected, the degree of difference between the predicted block and the source block is smaller than that of the source block. Has not been considered a problem so far.

However, this false detection of a motion vector is not a problem that can be ignored. The reason is that if a large number of irregular motion vectors appear in a region that originally moves uniformly, the amount of code required to encode the motion vector will increase significantly, and as a result, the compression ratio Decreases,
This is because the image quality is reduced.

[0008]

Recently, the motion vector detecting method shown in FIG. 2, which has been devised based on such recognition and semi-forcibly aligns adjacent motion vectors, has a large image quality at the time of video encoding. (199)
2008 IEICE General Conference D-11-50
Study of image quality improvement in MPEG2 video coding). In this method, when the motion vector of the encoding target block is obtained, n (block n in FIG. 1) is determined in advance.
And several blocks n-5 to n-1 and n + 1 to n + 5 (n-2, n-1, n + 1, n + 2, etc. in FIG. 1) located in front and behind in the coding order. A motion vector of a predicted block candidate (candidate of a motion vector) that falls within a difference value equal to or less than a value obtained by multiplying the minimum difference value by a constant is obtained, and a predicted block candidate and a motion vector of a neighboring block are aligned with each other. Is prioritized over the small degree of difference to determine the final motion vector.

In FIG. 2, motion vector candidates for block n-5: C, D, E, G,
H, I → E means that block n-5 may have been obtained by (i) the motion represented by vector C from one block, and (ii) represented by vector D from some other block. (Vi) may have been obtained by a motion represented by a vector I from another certain block, but finally consider the motion vector of a neighboring block. The result indicates that the result is obtained by the motion represented by the vector E. Of course, the other blocks n-4, n-3,...
The same applies to n,..., n + 4, and n + 5.

However, in this method based on the motion vector detection method shown in FIG. 2, in order to align the motion vector with the neighboring block, the motion vector cannot be determined unless the motion vector of the neighboring block is determined at the same time. Disrupts the regular pipeline processing of the unit. That is, if simultaneous determination with neighboring blocks is to be performed without inconsistency, simultaneous determination must be performed for all blocks that may have similar motion vectors, and usually one slice (image having m rows and n columns) is required. If it consists of blocks,
This is because it is necessary to determine a motion vector in units of one row).

For this reason, the processing after the motion detection of each block of the slice cannot be started until the motion detection processing has been completed up to the last block of the slice, and regular pipeline processing can be performed in block units. The hardware configuration is complicated as compared with the conventional processing. In addition, there is a drawback that depending on the contents of the image, a large number of motion vector candidates appear, and considerable hardware is required to select the most appropriate motion vector from the motion vectors in real time.

According to the present invention, motion detection is performed to reduce the difference between motion vectors between adjacent blocks to a necessary minimum without increasing the degree of difference between a motion detection target block and a prediction block in a reference image. It is an object of the present invention to realize a device with a simple hardware configuration that repeats the same processing for each processing cycle in block units.

[0013]

According to the present invention, not only a conventional motion detection block alone is used as a template to search in a reference image, but also an adjacent block (preceding adjacent block) located immediately before in the detection order. ) Or a block obtained by connecting the immediately adjacent block (subsequent adjacent block) and the block to be detected is used as a template to search the reference image. If the difference is slightly higher than that of the single template, Is to make the search result of the connected template preferentially selected so that the motion vectors are aligned between adjacent blocks. In addition, a motion vector closer to the motion vector of the preceding adjacent block is preferentially selected from the search results of the connected template, so that the motion vectors are more easily aligned between the adjacent blocks.

If the search for a linked template is performed directly and honestly, the amount of calculation is more than doubled. In order to avoid this, all the single difference degrees obtained by the motion detection of the adjacent block are held, and the single difference degree is obtained for the motion detection target, and at the same time, the obtained difference and the stored adjacent difference are stored. The difference degree of the connected template is obtained by adding the difference degree of the prediction block candidate giving the same motion vector of the block. This means that the search is not performed for each connected template.

With respect to the result of addition with the difference degree of the adjacent block, when the difference degree of the adjacent block is as small as that of the detection target, the difference degree of the connected template is smaller than the difference degree of the detection target block alone. Weighting is appropriately performed so that the evaluation position with the smaller degree of difference at the time of connection is preferentially selected.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS (First Embodiment) FIG. 3 shows an embodiment of the present invention in which an original block to be detected is equivalently connected to a preceding adjacent block as a search template. FIG. 3 is a block diagram of a first embodiment of FIG.

In this figure, reference numeral 10 denotes a necessary position of a search area of a reference image, which is cut out as a predicted block candidate P in a predetermined order for each block cycle, and calculates a difference A with respect to a single target block for motion detection; The difference A of the result and the cutout position are evaluated at the evaluation position (corresponding to the motion vector).
Is a difference degree evaluator output as A delay unit 20 delays the difference evaluation result by one block cycle and outputs the result as the difference Bf. Reference numeral 30 denotes a multiplier that takes the product of the difference Bf and a weight coefficient M input from the outside and outputs the product as the difference B'f.

Reference numeral 40 denotes an adder for calculating and outputting the sum of the difference A directly input from the difference evaluator 10 and the difference B'f. A multiplier 32 calculates a product of an input from the adder 40 and a weight coefficient N input from the outside, and outputs the result as a correction difference degree.

Reference numeral 50 denotes a comparison selector which compares the corrected difference with the difference A directly output from the difference evaluator 10, and outputs the smaller one as the final difference with respect to the evaluation position. . In each block cycle, if the final difference sent from the comparison selector 50 every time the evaluation position is updated is smaller than the difference held as the previous minimum value, the new minimum difference 60 The degree is a minimum value & evaluation position holder to be replaced with the evaluation position as a pair. Here, the block cycle is a processing cycle for evaluating all the degrees of difference at the positions requiring search with respect to one detection target block.

In this embodiment, every time a detection target block and a reference image are input, the difference degree evaluator 10 sequentially cuts out a block at a required position of the reference image as a prediction block candidate P of the detection target block, and performs detection. Calculate and output the difference degree for the target block. At the same time, the cutout position is output as an evaluation position.

The delay unit 20 delays the difference A received from the difference evaluator 10 by exactly one block cycle. As a result, the difference A directly supplied from the difference evaluator 10 and the difference B′f are input to the adder 40 at the same timing. Here, the difference B′f is the difference of the predicted block candidate Qf obtained with respect to the preceding adjacent block having the same evaluation position where the motion detection precedes by one block cycle, The difference Bf is obtained by delaying by one block cycle and multiplied by the weight coefficient M.

The adder 40 calculates the sum of these two degrees of difference. This sum is a difference degree with respect to the template in which the detection target block and the preceding adjacent block are linked, and is called a connection difference degree. In this state, the connection difference degree becomes larger than the single difference degree A. Therefore, multiplier 3
In step 2, the product is multiplied by the weight coefficient N so that the weight of the connection difference is equal to the difference A, and is output to the comparison selector 50 as the corrected connection difference.

In the comparison selector 50, the corrected connection difference from the multiplier 32 and the direct difference A from the difference evaluator 10 are calculated.
And outputs the smaller one (if equal, one of them) to the minimum value & evaluation position holder 60. In the minimum value & evaluation position holding unit 60, if the input from the comparison / selection unit 50 is smaller than the previously stored minimum difference in the same block cycle, the final difference of the input is determined. The corresponding evaluation position (difference evaluator 1
(Directly input from 0) as a new minimum difference and its evaluation position. Otherwise, the minimum difference and the evaluation position retain their previous values.

When one block cycle of the search is completed,
With respect to the detection target block, all the evaluations of the positions required for the search are completed, and the minimum value & evaluation position holder 60 stores the final minimum difference degree for the detection target block and the evaluation position corresponding thereto. can get. From the minimum value & evaluation position holding unit 60, the evaluation position giving the minimum degree of difference is output as the position of the prediction block as it is or converted into a motion vector for each block cycle.

Here, the portion composed of the multipliers 30 and 32 and the adder 40 is corrected by forming an averager in which the contribution of both the difference between the detection target block and the difference between the preceding adjacent block is variable. The connection difference degree is obtained, and if the weight coefficient M is set to "1" and the weight coefficient N is set to "1/2", the corrected connection difference degree becomes a simple average of both the difference degrees. If the weight coefficient N is set to a value smaller than "1/2", the corrected connection difference degree becomes smaller than the simple average, so that the difference degree Bf
Is as small as the difference A, the corrected connection difference is output to the minimum value & evaluation position holder 60 as the final difference instead of the difference A.

As is apparent from the above operation, in the present embodiment, in addition to the difference degree of the detection target block itself, the difference degree of the predicted block candidate Qf of the preceding adjacent block having the same evaluation position is used as the corrected connection difference degree. By evaluating at the same time,
Even if the difference A of the detection target block alone is equal to or slightly larger than the minimum difference A, an evaluation position with a small difference of the preceding adjacent block is preferentially selected.

Therefore, even for an image having a uniform brightness or pattern such as the sky, walls, and lawn, the motion vector is less likely to be violent than in the case of simply evaluating the difference of the motion detection target block alone. The motion vectors between them can be easily aligned.

As a matter of course, this effect works more effectively in a motion search for a sub-sampled image in which the detection accuracy tends to decrease and the motion vector tends to vary. Also, brightness,
In the case where the difference of the pattern or the like is clear and the source block of the detection target block is limited to one, the difference A becomes smaller than the connection difference and is effectively evaluated with the difference A. Therefore, there is no possibility that prediction cannot be made correctly unlike the case of always evaluating based on the degree of consolidation difference. further,
As is clear from the configuration and operation of the embodiment, the whole is always constituted by processing in units of blocks, and it is not necessary to wait for the motion vectors of neighboring blocks to be completed as in the conventional example. The consistency with the processing unit in units of blocks is high.

FIG. 4 is a diagram for explaining a situation in which processing is performed by pipeline processing in the first embodiment shown in FIG. In FIG. 4, (i) A _1K, A _2K
... Correspond to the degree of difference A shown in FIG.
_{1K ',} A _2K' ............ corresponds to the difference of Bf shown in FIG. 3, in response to _{(iii) MA 1K ', MA} 2K' differential degrees B'f ...... is illustrated in FIG. 3 I have. The difference degree evaluator 10 calculates (A _1K, P _1K ), (A
_{2K, P2K} ), ( _{A3K, P3K} ), ( _{A4K, P4K} ), (A
_5K, P _5K ) is output. The delay unit 20, the multiplier 30, the adder 40, and the multiplier 32 emit the output shown in each cycle. The comparison selector 50 outputs, for example, the smaller one of N (MA _1K ′ + A _2K ) and A _2K as shown in the figure. Minimum & Rating position retainer 60 also, as shown, and outputs the smaller, for example, B _2K and S _h2. Then, after one block cycle is completed, the evaluation position _Ph2 is output as it is or converted into a motion vector.

Second Embodiment FIG. 5 shows a search template according to the present invention in which a search template is constructed by equivalently connecting both a preceding adjacent block and a succeeding adjacent block in addition to a block to be detected. It is a block diagram of a 2nd example. The difference from the first embodiment shown in FIG. 3 is that one delay unit 20 is added, the original delay unit 20 is connected in series, the difference A is extracted from the preceding delay unit 20, and the subsequent delay unit is added. The difference Bf is derived from 20. The direct output from the difference evaluator 10 is drawn as the difference Bb, and the number of inputs is increased from “2” to “3” via the added multiplier 31. The input to the adder 41, the delay unit 20 'is added, and the minimum value & evaluation position holder 6
The three points are that the evaluation position input to 0 is delayed by one block cycle.

Due to these changes, the difference A becomes the difference B
b, one block cycle later than b, difference B
f has a relationship of the difference A delayed by one block cycle with respect to the difference A. At the same timing, assuming that the difference A is the difference between the prediction block candidate P and the detection target block, the difference Bf is the difference between the prediction block candidate Qf with respect to the preceding adjacent block and the difference Bb is the prediction with respect to the succeeding adjacent block. The difference is the difference between the block candidates Qb. Since the delay amount of each delay unit is set to exactly one block cycle, the evaluation positions of the predicted block candidates P, Qf, and Qb are all equal.

This is because the motion vector from the predicted block candidate P to the detection target block, the motion vector from the predicted block candidate Qf to the preceding adjacent block, and the motion vector from the predicted block candidate Qb to the succeeding adjacent block are:
In other words, it means that the detection target block, the preceding neighboring block, and the succeeding neighboring block are connected to each other, and the degree of difference with respect to the evaluation position at which a certain motion vector is given as a template is evaluated. .

The differences Bf and Bb are calculated by the multiplier 3
The product of the weighting factors Mb and Mc is calculated by using 0 and 31, and the difference A is input to the adder as it is. Multipliers 30, 3
1, 32 and the adder 41 are variable averagers of the input contribution. When the difference coefficients Bf and Bb are as small as the difference A by appropriately setting the weighting coefficients Mb, Mc and N. The value smaller than the difference A is the corrected connection difference,
It is input to the minimum value & evaluation position holder 60.

As a result, when the differences Bf and Bb are as small as the difference A, that is, when the difference between the preceding adjacent block and the succeeding adjacent block is as small as the difference between the detection target block and The corrected connection difference degree becomes smaller than the difference degree A for the detection target block alone, and remains in the minimum value & evaluation position holder with higher priority than in the case where it is not the case. In this second embodiment,
Since the size of the concatenated block becomes large, the variation of the motion vector can be suppressed more than in the first embodiment when the brightness and the pattern of the reference image are uniform. With the same effect, the front and rear vectors are more easily aligned than in the first embodiment.

FIG. 6 is a view for explaining a situation in which processing is performed by pipeline processing in the second embodiment shown in FIG. In FIG. 6, (i) C _1K, C _2K
... Correspond to the degree of difference Bb shown in FIG.
(Ii) C _1K ′ _, C _2K ′... Correspond to the difference A shown in FIG. 5, and (iii) C _1K ” _, C _2K ”... (Iv) MbC _1K "corresponding to the degree of difference Bf
_, MbC _2K "... Correspond to the degree of difference B'f shown in FIG. 5, and (v) McC _1K, McC _2K ... Correspond to the degree of difference B'b shown in FIG. The values output by the degree evaluator 10 or the minimum value & evaluation position holder 60 are shown in FIG.
6 is shown in the frame in FIG.

(Third Embodiment) In the first and second embodiments, although the motion vector is less likely to vary, when the connected template overlaps the boundary of the area, the corrected connection difference degree increases, and the evaluation of the connected template is increased. There is a problem that the position is hardly left and the effect of aligning the motion vectors is reduced.

To solve this problem, FIG. 7 shows a connection between a preceding adjacent block and a detection target block, a connection between a subsequent adjacent block and a detection target block, a connection between both adjacent blocks and a detection target block, FIG. 13 is a block diagram of a third embodiment in which four types of templates of a detection target block alone are simultaneously evaluated. As shown in this figure, the difference from the second embodiment is that an adder 42 and a multiplier 34 for evaluating the degree of difference between the preceding adjacent block and the detection target block with respect to the connected template are added. Adding an adder 43 and a multiplier 35 for evaluating the degree of difference between the succeeding adjacent block and the detection target block with respect to the connected template; and expanding the comparison selector 51 from two inputs to four inputs. Is a point.

As is apparent from this configuration, by appropriately setting the weighting factors N2 and N3 to be applied to the added multipliers 34 and 35, the boundary of the area overlaps only one adjacent block, and the adjacent block on the boundary side is overlapped. Even if the corrected connection difference degree of the template including is large, the corrected connection difference degree of the connection template with the adjacent block on the opposite side to the boundary remains easily reduced. Therefore, even when the boundary overlaps with the adjacent block, the motion vector of the detection target block can be easily aligned with the motion vector of the adjacent block on the opposite side of the boundary.

FIG. 8 is a diagram for explaining a situation in which processing is performed by pipeline processing in the third embodiment shown in FIG. In FIG. 8, (i) C _1K, C _2K
... Correspond to the degree of difference Bb shown in FIG.
(Ii) C _1K ′ _, C _2K ′... Correspond to the degree of difference A shown in FIG. 7, and (iii) C _1K ” _, C _2K ”.
(Iv) MbC _1K ” _, M
bC _2K "... corresponds to the degree of difference B'f shown in FIG.
(v) McC _1K, McC _2K ... correspond to the degree of difference B′b shown in FIG. The values output by the difference degree evaluator 10 or the minimum value & evaluation position holder 60 are shown in FIGS.
8 is shown in the frame in FIG.

(Fourth Embodiment) FIG. 9 shows a fourth embodiment of the present invention. The difference from the first embodiment is that a difference degree modification mechanism including a multiplier 33, a holder 65, a distance calculator 70, and a conversion table 80 is added.

In this difference degree modification mechanism, a distance calculator 70 for calculating a distance between a position of a prediction block (corresponding to a motion vector) of a preceding adjacent block held in a holder 65 and an evaluation position of a detection target block. Is provided to the multiplier 33 via the conversion table 80.

If the distance falls below a predetermined threshold, the conversion table 80 is set so that the smaller the distance value is, the smaller the value is given to the multiplier. , The final difference degree is likely to be small, and the prediction block candidate at a position close to the motion vector of the preceding adjacent block is easily selected as the prediction block.

With this function, the motion vectors of adjacent blocks can be more easily aligned than in the first embodiment. Of course, this embodiment incorporates the second and third embodiments in which the difference between the linked template with the preceding or succeeding adjacent block is evaluated in addition to the difference between the detection target blocks. It is possible to incorporate a weighting factor switching mechanism, thereby making it easier to align the motion vectors of adjacent blocks.

FIG. 10 is a diagram for explaining a situation in which processing is performed by pipeline processing in the fourth embodiment shown in FIG. In FIG. 10, (i) A _1K, A
_2K ... Correspond to the degree of difference A shown in FIG.
i) A _1K ′ _, A _2K ′... are the degrees of difference Bf shown in FIG.
(Iii) MA _1K ′, MA _2K ′... Correspond to the difference B′f shown in FIG. The values output by the difference degree evaluator 10 or the minimum value & evaluation position holder 60 are shown in FIG.
As in the case of FIG. 6, FIG. 6 and FIG. 8, it is shown in the frame in FIG.

In the above four embodiments, weighting or correction for each difference is realized by multiplying by a weight coefficient using a multiplier for easy understanding. However, heavy use of the multiplier causes an increase in hardware scale. When priority is given to the hardware scale, there is a method in which a subtracter is used instead of a multiplier to subtract one power of 2 from the input difference degree. With this method, an arbitrary weighting factor cannot be given, but the hardware scale can be significantly reduced.

Further, in the above four embodiments, the reference of the difference degree of the prediction block which gives the same motion vector to the adjacent block as that of the detection target block is also performed.
In order to make it easy to understand, they are realized by delaying one block cycle or two block cycles using a delay device.

Indeed, this method is suitable for a case where the search range for the detection target block is always fixed. However, in a search in which the search range is adaptively shifted according to the motion of the image, the relationship between the detection target block and the search area differs for each block cycle, so even if the search range is scanned in a fixed order. In a delay unit having a fixed delay time, motion vectors cannot refer to the same degree of difference.

In this case, the correspondence table between the evaluation position and the degree of difference can be realized by a memory, and the difference degree whose evaluation position is the same as that of the detection target block is read from the correspondence table of the adjacent block. However, in this method, when the search range is different from the adjacent block, the same evaluation position may not exist in the correspondence table of the adjacent block. In this case, the maximum degree of difference is read out, and the linked correction degree of difference at that position is made to greatly exceed the degree of difference alone, so as not to remain as the final degree of difference. By doing so, only when the same evaluation position exists in the search for the adjacent block,
It is possible that the connection correction difference degree can be smaller than the single difference degree.

In the above description, the configuration of the first to fourth embodiments as an apparatus for realizing the present invention has been described. However, the present invention can be realized not only by the device, but also by the block diagram shown in FIG. 3 for realizing the first embodiment, the block diagram shown in FIG. 5 for realizing the second embodiment, and FIG. For each of the block diagram shown in FIG. 7 for realizing the third embodiment and the block diagram shown in FIG. 9 for realizing the fourth embodiment, the flow of the processing shown in each block diagram will be described as it is. It is possible to rewrite the shape, and such a flowchart and a flow of the process can be regarded as a method embodying the motion detection method according to the present invention. The present invention also includes the motion detection method within the scope of rights, and is described in claims 6 to 10.

Further, the motion detection method can be expressed by a program executable by the data processing device. Therefore, the present invention also includes a recording medium on which the program of the motion detection method is recorded, as described in claims 11 to 15.

[0051]

As described above, according to the present invention,
If the brightness or pattern of the image is uniform, the evaluation is performed using a template in which the adjacent block is connected to the detection target block. In such a case, the motion vector is less likely to vary, and the motion vector between the adjacent blocks is reduced. Also become easier. Also, when the difference in brightness and pattern is clear and the source block is specified as one, the block to be detected is evaluated using the template of the detection target block alone. As described above, there is no concern that detection of a prediction block is likely to fail. Furthermore, it can be realized by processing in units of blocks, and is highly compatible with devices realized by pipeline processing in units of blocks, such as video coding. Thus, the present invention provides
Since the increase in the size and complexity of the hardware can be minimized and the motion vectors between the adjacent blocks can be easily aligned, it is extremely useful for economically realizing a high-performance motion detector.

[Brief description of the drawings]

FIG. 1 is an example of a conventional standard motion search.

FIG. 2 is a diagram illustrating determination of a motion vector performed after all motion vector candidates of neighboring blocks are available.

FIG. 3 is a block diagram of a first embodiment of the present invention for evaluating both a detection target block alone and a block obtained by linking a preceding adjacent block to the detection target block as a search template;

FIG. 4 is a diagram illustrating a situation in which processing is performed by pipeline processing for the first embodiment shown in FIG. 3;

FIG. 5 is a block diagram of a second embodiment of the present invention which evaluates both a detection target block alone and a block obtained by connecting a preceding adjacent block and a succeeding adjacent block as a search template.

FIG. 6 is a diagram illustrating a situation in which processing is performed by pipeline processing in the second embodiment illustrated in FIG. 5;

FIG. 7 evaluates four types of search templates: a detection target block alone, a connection of a preceding adjacent block, a connection of a subsequent adjacent block, and a connection of a preceding adjacent block and a subsequent adjacent block. FIG. 11 is a block diagram of a third embodiment of the present invention.

FIG. 8 is a diagram illustrating a situation in which processing is performed by pipeline processing in the third embodiment illustrated in FIG. 7;

FIG. 9 is a block diagram of a fourth embodiment of the present invention for making it easy to align a motion vector to be detected with a motion vector of a preceding adjacent block.

FIG. 10 is a diagram for explaining a situation in which processing is performed by pipeline processing in the fourth embodiment shown in FIG. 9;

[Explanation of symbols]

 Reference Signs List 10 difference degree evaluator 20, 20 'delay unit 30, 31, 32, 33, 34, 35 multiplier 40, 41, 42, 43 adder 50, 51 comparison selector 60 minimum value & evaluation position holder 65 holder 70 distance calculator 80 conversion table

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Mitsuro Ikeda 3-19-2 Nishi-Shinjuku, Shinjuku-ku, Tokyo F-term in Nippon Telegraph and Telephone Corporation (reference) 5C059 KK19 NN10 NN11 NN28

Claims (15)

[Claims] 1. A motion detection for finding, as a prediction block, a block which can be regarded as a movement source from another reference image from a small difference degree of a block in an image, and obtaining a motion vector from the prediction block to the block in the image. The motion vector from the prediction block candidate P to the detection target block is smaller than the single degree of difference A of the prediction block candidate P of the reference image with respect to the detection target block in the determination as to whether or not to be the prediction block. Means for giving the same motion vector to a preceding neighboring block adjacent in the detection order and prioritizing both the difference B and the difference A of the intra prediction block candidate Q with respect to the preceding neighboring block. A motion detector, characterized in that: 2. A motion vector identical to a motion vector from a prediction block candidate P obtained when a preceding adjacent block is processed as a detection target to a detection target block is determined as a difference B with respect to the preceding adjacent block. 2. The motion detector according to claim 1, wherein a difference degree of a predicted block candidate Q with respect to a preceding adjacent block is used. 3. The method according to claim 1, wherein a constant multiple of the average value of the difference A and the average value of the difference B is used as the smallness of both the difference A and the difference B. Motion detector. 4. In the determination of a prediction block, the same motion vector as the motion vector given by the prediction block candidate P of the detection target block to the detection target block candidate is given to the preceding adjacent block in addition to the difference A. Difference degree Bf of predicted block candidate Qf with respect to the preceding adjacent block
And a predicted block candidate Qb that gives the same motion vector to a succeeding adjacent block located immediately after in the detection order.
Means for prioritizing the smaller of the three different degrees of difference Bb with respect to the succeeding adjacent block than the single difference A of the predicted block candidate with respect to the detection target block. 2. The motion detector according to 1. 5. The motion according to claim 1, further comprising means for giving priority to the proximity of the motion vector of the detection target block to the motion vector of the preceding adjacent block, in addition to the small difference degrees A and B. Detector. 6. For a block in an image, a block that can be regarded as a movement source is found as a prediction block from another reference image because of a small degree of difference, and motion estimation for obtaining a motion vector from the prediction block to the block in the image is performed. In the method, when determining whether or not to be a prediction block, a motion vector from the prediction block candidate P to the detection target block is smaller than the single degree of difference A of the prediction block candidate P of the reference image with respect to the detection target block. The same motion vector as that given in the detection order is given to the adjacent preceding block adjacent to the preceding adjacent block in the reference image, and the difference B between the preceding adjacent block and the small difference A to the preceding adjacent block is prioritized. A motion detection method characterized in that: 7. When obtaining a degree of difference B with respect to a preceding adjacent block, a motion vector identical to the motion vector from the prediction block candidate P obtained when the preceding neighboring block is processed as a detection target to the detection target block is obtained. 7. The motion detection method according to claim 6, wherein a degree of difference of a prediction block candidate Q given to an adjacent block with respect to a preceding adjacent block is obtained. 8. The method according to claim 6, wherein a constant multiple of the average value of the difference A and the average value of the difference B is used as the smallness of both the difference A and the difference B. Motion detection method. 9. In determining a prediction block, a prediction block candidate P of a detection target block is added to the difference degree A.
Is given to the preceding neighboring block the same motion vector as the motion vector given to the detection target block candidate, and the difference Bf of the predicted block candidate Qf from the preceding neighboring block, and the same motion vector is located immediately after in the detection order. Of the three different degrees of difference Bb of the predicted block candidate Qb given to the succeeding adjacent block to be performed with respect to the succeeding adjacent block is smaller than the difference A alone of the difference A of the predicted block candidate to the detection target block. 7. The motion detecting method according to claim 6, wherein priority is given also to the motion detection. 10. The motion according to claim 6, wherein the proximity of the motion vector of the detection target block to the motion vector of the preceding adjacent block is prioritized in addition to the small difference degrees A and B. Detection method. 11. A motion detection for finding, as a prediction block, a block which can be regarded as a movement source from another reference image from a small difference degree of a block in an image, and obtaining a motion vector from the prediction block to the block in the image. In a recording medium on which a program corresponding to the method is recorded, when determining whether or not to be a predicted block, the predicted block candidate P of the reference image is smaller than the difference degree A of the detected block with respect to the detection target block. Both the difference B and the difference A of the prediction block candidate Q in the reference image with respect to the preceding adjacent block which gives the same motion vector as the motion vector from P to the detection target block in the detection order to the preceding adjacent block Program corresponding to a motion detection method characterized by giving priority to smallness Recorded recording medium. 12. When obtaining a difference B with respect to a preceding adjacent block, the same motion vector as a motion vector from the prediction block candidate P obtained when the preceding neighboring block is processed as a detection target to the detection target block is used. 12. A recording medium storing a program corresponding to the motion detection method according to claim 11, wherein a degree of difference of a predicted block candidate Q given to an adjacent block with respect to a preceding adjacent block is obtained. 13. The method according to claim 11, wherein a constant multiple of the average value of the difference A and the average value of the difference B is used as the smallness of both the difference A and the difference B. A recording medium on which a program corresponding to the motion detection method is recorded. 14. In determining a prediction block, the same motion vector as the motion vector given by the prediction block candidate P of the detection target block to the detection target block candidate is given to the preceding adjacent block in addition to the difference A. The difference Bf of the prediction block candidate Qf with respect to the preceding adjacent block and the difference Bb of the prediction block candidate Qb with respect to the succeeding adjacent block immediately after the same motion vector in the detection order are given to the succeeding block. 3
12. A program corresponding to the motion detection method according to claim 11, wherein the smaller of the two degrees of difference is prioritized over the smaller of the degree of difference A alone of the prediction block candidate with respect to the detection target block. Recording medium. 15. The motion according to claim 11, wherein the proximity of the motion vector of the detection target block to the motion vector of the preceding adjacent block is prioritized in addition to the small difference degrees A and B. A recording medium on which a program corresponding to the detection method is recorded.