JPH07123406A - Motion vector detector - Google Patents

Abstract

PURPOSE:To improve the accuracy of a motion vector. CONSTITUTION:This device is provided with a motion vector classifying circuit 8 for comparing a certain motion vector with the surrounding motion vector for each motion vector as a prescribed block unit to be detected by a motion vector detection circuit 5 and for classifying the motion vector of the block unit based on this compared result. Based on the magnitude difference to be provided as the compared result of both motion vectors, the reliability of the motion vector for the unit of a block is decided for each block and corresponding to this decided result, a motion vector decision circuit 6 decides any specified motion vector. Then, the motion vector of a block unit to cause the generation of any error at the specified motion vector is invalidated, and only the other valid motion vector of the block unit is used for deciding this specified motion vector. Thus, the detection accuracy of the motion vector is improved.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a motion vector detecting device, and particularly to a motion vector detecting device adapted to detect a motion vector from an image signal.

[0002]

2. Description of the Related Art At present, as a motion vector detecting method used in an image coding apparatus, an image blur correcting apparatus, etc., a method called a correlation method based on a correlation operation or a block matching method is known. For the matching calculation, see Morio Onoue et al. "Information Processing Vol.17, No.7,
p634-640 July 1976 ".

In the block matching method described above, first, an input image signal is divided into a plurality of blocks having an appropriate size (for example, 8 pixels × 8 lines). Next, the difference between the pixel signal of the current field (or frame) and the pixel signal of a certain range of the previous field (or frame) is calculated in block units, and the sum of the absolute values of this difference is calculated to be the minimum value. Search for a block of fields (or frames). Then, the relative displacement of the blocks searched in this way is detected as a motion vector in block units.

Next, an example of a conventional motion vector detecting device to which this block matching method is applied will be described with reference to the drawings. FIG. 14 is a block diagram showing the configuration of an image blur prevention device to which a conventional motion vector detection device is applied.

In FIG. 14, first, an image signal to be a motion vector detection target is input to the field (or frame) memory 1, and the input image signal is temporarily stored until the following processing is performed. It

On the other hand, the image signal to be the detection target of the motion vector is also input to the spatial frequency filter 2. Then, the spatial frequency filter 2 removes the low-frequency component and the high-frequency component from the input image signal, thereby extracting the spatial-frequency component useful for detecting the motion vector.

The image signal subjected to the predetermined filtering processing by the spatial frequency filter 2 is supplied to the correlation calculation circuit 3 and the memory 4. Further, the image signal supplied to the memory 4 is subjected to delay processing for one field period and then supplied to the correlation calculation circuit 3. This allows
The image signal of the current field and the image signal of the previous field are supplied to the correlation calculation circuit 3.

Then, the correlation calculation circuit 3 performs the correlation calculation between the image signal of the current field and the image signal of the previous field in block units according to the block matching method, and the correlation value obtained as a result of this calculation is the motion vector. It is supplied to the detection circuit 5.

Next, the motion vector detection circuit 5 calculates the correlation value based on the correlation value given from the correlation calculation circuit 3.
A block-by-block motion vector is detected. Specifically, the block in the previous field having the minimum correlation value is searched for, and the relative shift of this block is detected as a motion vector in block units.

The block-by-block motion vector thus detected is supplied to the motion vector determination circuit 6. Then, the motion vector determination circuit 6 determines the motion vector of the entire field from the block-by-block motion vector. Specifically, the median value or the average value of the block-by-block motion vectors is determined as the motion vector of the entire field.

The motion vector of the entire field thus determined is supplied to the memory read control circuit 7.
Then, the memory read control circuit 7 controls the read position of the image signal stored in the memory 1 so that the shake of the image is canceled according to the motion vector of the entire field. By performing the above-described control, the image signal with the shake prevented is read from the memory 1 and output to the next stage.

[0012]

However, the motion vector detected by the above-described motion vector detection circuit 5 in block units depends on the image state of the target block. Therefore, a reliable motion vector is detected when the image state of the target block is good, but a motion vector with a large error is often detected when the image state of the target block is not good.

That is, as shown in FIG. 2 (a), a subject having few characteristic patterns in an image, a subject having many patterns having similar characteristics, and a subject having a strong correlation in a specific direction. Are called poor subjects. When attempting to detect a motion vector for such a weak subject, a motion vector with a large error may be detected as shown in FIG.

Further, as shown in FIG. 5A, when an attempt is made to detect a motion vector for an image in which a subject is moving, as shown in FIG. 5B, the motion vector differs from the motion vector caused by the shake. Motion vectors with different directions and magnitudes were sometimes detected. Therefore, the area of the motion vector detected by the motion vector detection circuit 5 may be divided into the area of the motion vector generated by the shake and the area of the motion vector generated by the motion of the subject.

In such cases, when the motion vectors detected in all the blocks in the image are used to determine the motion vector of the entire field as in the conventional case,
There is a problem in that a correct motion vector cannot be obtained and the motion vector detection accuracy is significantly deteriorated.

The present invention has been made to solve such a problem, and an object of the present invention is to provide a motion vector detection device and method capable of improving the motion vector detection accuracy.

[0017]

A motion vector detecting apparatus according to the present invention is based on a correlation calculation means for performing a correlation calculation between time-sequential images and a correlation value calculated by the correlation calculation means. Motion vector detecting means for detecting a motion vector, motion vector determining means for determining a specific motion vector based on the motion vector detected by the motion vector detecting means, and motion vector detected by the motion vector detecting means For each, a certain motion vector is compared with another motion vector, the motion vector is classified based on the difference between both motion vectors obtained as a result of this comparison, and the classification result is derived to the motion vector determination means. And a motion vector classification means.

In the motion vector detecting device having the above-described configuration, the correlation calculation, the motion vector detection, and the motion vector classification may be performed in a predetermined block unit. Further, the motion vector determining means determines the specific motion vector using the block-by-block motion vector of the valid classification among the block-by-block motion vectors classified by the motion vector classifying means. May be.

Another feature of the present invention is that the binarizing means for binarizing the image signal, and the binarizing means
Correlation calculation means for performing a correlation calculation between time-sequential binary images composed of image signals subjected to the binarization process, and a motion vector for detecting a motion vector based on the correlation value calculated by the correlation calculation circuit. Based on the detection means and the motion vector detected by the motion vector detection means,
In the motion vector determination means for determining a specific motion vector, the correlation value detection means for detecting the spatial distribution of the correlation values calculated by the correlation calculation means, and the spatial distribution of the correlation values detected by the correlation value detection means Motion vector evaluation means for evaluating the reliability of the motion vector detected by the motion vector detection means based on the change and absolute value of the correlation value gradient and deriving the evaluation result to the motion vector determination means. To do.

[0020]

Since the present invention comprises the above technical means, the reliability of the motion vector detected by the motion vector detecting means is judged, and the motion vector which causes an error in the specific motion vector to be detected is invalid. Then, only other useful motion vectors will be used to determine the specific motion vector.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a shake prevention device to which a motion vector detection device according to a first embodiment of the present invention is applied. In the first embodiment, it is possible to discriminate between a motion vector having a large error and a motion vector having a small error, which are detected with respect to a poor subject as shown in FIG.

In FIG. 1, first, an image signal s1 as a motion vector detection target is input to a field (or frame) memory 1 and temporarily stored until the following processing is performed.

On the other hand, the image signal s1 to be the motion vector detection target is also input to the spatial frequency filter 2. Then, the spatial frequency filter 2 removes the low-frequency component and the high-frequency component from the input image signal, thereby extracting the spatial-frequency component useful for detecting the motion vector.

Then, the image signal s subjected to the above-mentioned filtering processing by the spatial frequency filter 2
2 is supplied to the correlation calculation circuit 3 and the memory 4. Further, the image signal s2 supplied to the memory 4 is delayed by one field period and then supplied to the correlation calculation circuit 3. As a result, the correlation calculation circuit 3 is supplied with the image signal s2 of the current field and the image signal s3 of the previous field.

Then, the correlation calculation circuit 3 makes the image signal s of the current field according to the block matching method.
The correlation calculation between 2 and the image signal s3 of the previous field is performed in block units, and the correlation value s4 obtained as a result of this calculation is supplied to the motion vector detection circuit 5.

Next, the motion vector detecting circuit 5 detects the block-by-block motion vector s5 based on the correlation value s4 provided from the correlation calculating circuit 3. Specifically, the block in the previous field in which the correlation value s4 is the smallest is searched, and the relative shift of this block is detected as the motion vector s5 in block units.

The block-by-block motion vector s5 thus detected is supplied to the motion vector determination circuit 6 and the motion vector classification circuit 8. Then, in the motion vector classification circuit 8, the motion vector of each block is compared with the motion vectors of the surrounding blocks to generate classification information s6 in block units as described later. The generated classification information s6 is supplied to the motion vector determination circuit 6.

Next, the motion vector determination circuit 6 determines the motion vector s7 of the entire field by using only the motion vector of the block which is the effective area according to the classification information s6. Specifically, the median value or the average value of the target block-based motion vectors is determined as the motion vector s7 of the entire field.

The motion vector s7 of the entire field thus determined is supplied to the memory reading control circuit 7. Then, by this memory read control circuit 7,
The reading position of the image signal stored in the memory 1 is controlled so that the shake of the image is canceled according to the motion vector s7. By performing the above control, the shake-prevented image signal s8 is read from the memory 1 and output to the next stage.

Next, a method of classifying motion vectors performed by the motion vector classifying circuit 8 described above will be described with reference to FIG.
And it demonstrates using FIG. Note that FIG. 2 is a diagram for explaining a manner in which a block-by-block motion vector is detected from an image input to the motion vector detection device and classification information is generated based on the block-by-block motion vector. 3 is a flowchart showing the operation of the motion vector classification circuit 8.

FIG. 2A shows a state in which an image as a motion vector detection target is divided into a plurality of blocks. Further, FIG. 2B shows an example of a block-unit motion vector detected for the image as shown in FIG. As is clear from FIG. 2B, a motion vector (motion vector facing the upper right direction) caused by camera shake is detected in a large number of blocks.

However, as shown in the image of FIG. 2 (a), there are few characteristic patterns such as the sky and walls, and in the case of a subject who is not good at having many patterns having similar characteristics such as a handrail, As shown in FIG. 2B, an error vector in a direction different from the motion vector caused by camera shake may be detected.

By the way, when detecting a motion vector for such a poor subject, there is a feature that a motion vector similar to the error vector is not detected around the error vector as described above. Therefore, the motion vector classification circuit 8 in the present embodiment utilizes such a feature to distinguish between a motion vector due to camera shake and an error vector due to a poor subject, and to generate predetermined classification information.

That is, the motion vector classification circuit 8 first obtains a block-unit motion vector as shown in FIG. 2B detected by the motion vector detection circuit 5 in step P1 of FIG.

Next, the motion vector classification circuit 8 calculates an evaluation value for determining whether or not the motion vector of a certain block is an error vector in steps P2 to P5. For example, when calculating an evaluation value for a block indicated by diagonal lines in FIG. 2B, this block is compared with eight surrounding blocks, and the evaluation value is calculated based on the result of this comparison. calculate.

Hereinafter, of the nine blocks surrounded by the thick line in FIG. 2B, the central block for which an evaluation value is to be obtained is called an evaluation block. The eight blocks around this evaluation block are called comparison blocks.

To calculate the evaluation value of this evaluation block, the motion vector classification circuit 8 first calculates the discriminant value of the evaluation block based on (Equation 1) in step P2. Discrimination value = | evaluation vector−comparison vector | (1 expression) Here, the evaluation vector is a motion vector in the evaluation block, and the comparison vector is a motion vector in a certain one comparison block.

Then, the motion vector classification circuit 8 compares the discriminant value of the evaluation block with a predetermined constant A in step P3 to determine whether or not the following conditional expression is satisfied. Discriminant value <constant A (Equation 2) Then, when this conditional expression is satisfied, it is considered that the evaluation vector as the comparison target and the comparison vector are similar,
In step P4, 1 is added to the evaluation value.

On the other hand, when the conditional expression shown in (2) is not satisfied, it is considered that the evaluation vector to be compared and the comparison vector are not similar to each other, so the process of step P4 is not performed and step P5 is performed. Proceed to.

In step P5, the motion vector classification circuit 8 confirms whether or not the comparison processing between one evaluation block and all the comparison blocks around it has been completed. When it is confirmed that all of the comparison processing is completed, the process proceeds to step P6, the evaluation value calculated as described above is compared with a predetermined constant B, and the conditional expression shown in the following (3 expression) is obtained. It is determined whether or not it is satisfied. Evaluation value> Constant B ... (3 expressions)

When the conditional expression (3) is satisfied, it is determined that the evaluation vector is not an error vector due to the subject that is not good, and the process proceeds to step P7 to classify "1" indicating that it is an effective block. It is set as information s6. On the other hand, when the conditional expression (3) is not satisfied, it is determined that the evaluation vector is an error vector, and step P
In step 8, "0" indicating an invalid block is set as the classification information s6.

Next, the motion vector classification circuit 8 confirms in step P9 whether or not the classification processing as described above has been completed for all blocks in the image, and if there is an unprocessed block, step P2. Return to. Then, steps P2 to P are performed until classification processing is performed on all blocks.
The process of 9 is repeated.

An example of the block-based classification information s6 obtained as a result of executing the classification processing on all blocks in this manner is shown in FIG. 2 (c). In FIG. 2C, the block indicated by “1” indicates that it is a valid block, and the block indicated by “0” indicates that it is an invalid block.

The classification information s6 as shown in FIG. 2 (c) is supplied to the motion vector determination circuit 6 in FIG.
When the motion vector determination circuit 6 determines the motion vector of the entire field, this classification information s6
Only the motion vector of the effective block is used based on

Next, a second embodiment of the present invention will be described with reference to the drawings. FIG. 4 is a block diagram showing the configuration of a shake prevention device to which the motion vector detection device of this embodiment is applied. In FIG. 4, the same parts as those in the motion vector detecting device shown in FIG. 1 are designated by the same reference numerals, and overlapping description will be omitted.

In the second embodiment, the motion vector classification circuit 8 in the first embodiment described above is replaced with a motion vector classification circuit 9 having a different classification method, as shown in FIG. The motion vector in various directions detected with respect to the moving subject can be discriminated.

A method of classifying each block by the motion vector classifying circuit 9 will be described below with reference to FIGS. 5 and 6. Note that FIG. 5 is a diagram for explaining how to detect a block-by-block motion vector from an image input to the motion vector detection device and generate classification information based on the block-by-block motion vector. Also,
FIG. 6 is a flowchart showing the operation of the motion vector classification circuit 9.

FIG. 5A shows a state in which an image to be a motion vector detection target is divided into a plurality of blocks. Further, FIG. 5B shows an example of a block-unit motion vector detected for an image as shown in FIG. As is clear from FIG. 5B, motion vectors in different directions are detected from the car moving to the left, the bus moving to the right, and the background in the stationary state.

Therefore, in the motion vector classification circuit 9 in this embodiment, a certain specific motion vector is selected as a reference vector, and this reference vector is compared with the motion vector of each block to determine each block in the image. The areas having motion vectors in different directions can be classified.

That is, the motion vector classification circuit 9 first obtains the block-by-block motion vector as shown in FIG. 5B detected by the motion vector detection circuit 5 of FIG. 4 in step P1 of FIG. get.

Then, the motion vector classification circuit 9 selects a certain reference vector to be compared in step P2. This reference vector may be, for example, an average value of motion vectors in all blocks, or may be a motion vector of any one block selected from all blocks as a representative.

Next, in step P3, the motion vector classification circuit 9 calculates the discriminant value of a certain evaluation block based on the following (Equation 4). Discrimination value = evaluation vector-reference vector (4) Further, the motion vector classification circuit 9 compares the discrimination value of this evaluation block with a predetermined constant A in step P4,
It is determined whether or not the conditional expression (5) is satisfied. Judgment value <Constant A ... (Equation 5)

If the conditional expression (5) is satisfied, the process proceeds to step P5, and "1" indicating the area of the first motion vector is set as the classification information. On the other hand, if the conditional expression (5) is not satisfied, the process proceeds to step P6, and the discriminant value and the predetermined constant B (> A) are compared to determine whether the conditional expression shown in the following (6) is satisfied. Determine whether or not. Judgment value <Constant B ... (6 expressions)

If this conditional expression is satisfied, the process proceeds to step P7, where "2" indicating the area of the second motion vector is set as classification information. If this conditional expression is not satisfied, the process proceeds to step P8 and the third
“3” indicating that the area is a motion vector area of is set as classification information.

Next, the motion vector classification circuit 9 confirms in step P9 whether or not the above classification processing has been completed for all blocks, and if there is an unprocessed block, the process returns to step P3. Then, the processes of steps P3 to P9 are repeated until the classification process is performed on all the blocks.

FIG. 5C shows an example of the classification information in block units obtained as a result of executing the classification processing on all blocks in this way. In FIG. 5C, the block indicated by “1” is the area of the first motion vector. Further, the block indicated by “2” indicates that it is the area of the second motion vector, and the block indicated by “3” indicates that it is the area of the third motion vector.

The classification information s10 as shown in FIG. 5C is supplied to the motion vector determination circuit 6 in FIG. Then, when the motion vector determination circuit 6 determines the motion vector of the entire field, based on this classification information s10, an area in which the image is to be made stationary among the areas of the first to third motion vectors. Only the motion vector (for example, the background area in FIG. 5A) is used.

In the present embodiment, the case where each block in the image is classified into the first to third regions by comparing the discriminant value of the evaluation block with the two constants A and B will be taken as an example. Although described, the present invention is not limited to this.

That is, the discriminant value of the evaluation block is set to 1
Each block in the image may be divided into two regions by comparing the two constants. Further, the discriminant value of the evaluation block is compared with three or more constants, or the motion vector of the block unit is divided into the X-direction component and the Y-direction component of the XY coordinates, and the discriminant value calculated for each and a predetermined constant It is also possible to classify each block in the image into more areas by comparing

It is also possible to use the classification method of each block according to the first embodiment and the classification method according to the second embodiment together.

Next, FIG. 7 is a flow chart showing the main operation of the motion vector detecting apparatus according to the above-mentioned first or second embodiment. In FIG. 7, first, in step P1, with respect to the image signal for which the motion vector is to be detected, the correlation calculation between the image signal of the current field and the image signal of the previous field is performed in block units according to the block matching method.

Then, in step P2, this step P
The motion vector in block units is detected based on the correlation value in block units obtained by the correlation calculation of 1. Specifically, the block in the previous field having the minimum correlation value is searched for, and the relative shift of this block is detected as a motion vector in block units.

In step P3, the same processing as the processing shown in the flowchart of FIG. 3 or the flowchart of FIG. 6 is performed based on the thus detected block unit motion vector, and the block unit as described above is executed. Classification information is generated.

Then, in step P4, the motion vector of the entire field is determined by using only the motion vector of the block which is regarded as the effective area by the generated classification information. Specifically, the median value or the average value of the target block-based motion vectors is determined as the motion vector of the entire field.

Next, a third embodiment of the present invention will be described with reference to the drawings. FIG. 8 is a block diagram showing the configuration of a shake prevention device to which the motion vector detection device according to the third embodiment is applied. In addition, in FIG.
The same parts as those in the motion vector detecting device shown in FIG.

In FIG. 8, the image signal s2 subjected to a predetermined filtering process by the spatial frequency filter 2.
Is binarized at zero level by the binarization circuit 10.
The binary image signal s11 binarized by the binarization circuit 10 is supplied to the correlation calculation circuit 3, and the current field image signal s11 and the previous field image signal s12 supplied from the memory 4 are combined. Correlation calculation between the two is performed.

The correlation value s13 between the current field and the previous field obtained as a result of the calculation by the correlation calculation circuit 3 is supplied to the motion vector detection circuit 5 and the correlation value detection circuit 11. The motion vector detection circuit 5 detects the block-based motion vector s14 based on the correlation value s13.

Specifically, the motion vector detection circuit 5 searches for a block in the previous field in which the above-mentioned correlation value s13 is the minimum, and the relative shift of this block is detected as a motion vector s14 in block units. . The motion vector s14 in block units thus detected is supplied to the motion vector evaluation circuit 12.

On the other hand, in the correlation value detection circuit 11, the minimum value of the correlation value given from the correlation calculation circuit 3 (the minimum correlation value searched by the above-mentioned motion vector detection circuit 5),
As shown in FIG. 9, this minimum correlation value and the correlation value of the surrounding positions that are apart from each other by a predetermined pixel centered on the position where the minimum correlation value is present are detected. Then, these detected correlation values s15 are
It is supplied to the motion vector evaluation circuit 12.

Next, the motion vector evaluation circuit 12 converts the block-by-block motion vector s14 detected by the motion vector detection circuit 5 into the correlation value detection circuit 11
Minimum correlation value detected by and the correlation value s around it
It is evaluated based on 15. The contents of this evaluation will be described later.

The motion vector s16 in block units evaluated in this way is supplied to the motion vector determination circuit 6, and the motion vector s16 in block units is given.
Based on the above, the motion vector s17 of the entire field is determined. Specifically, the median value or the average value of the given block-by-block motion vectors is determined as the motion vector s17 of the entire field.

Next, the method of evaluating the motion vector by the motion vector evaluation circuit 12 will be described with reference to the flowchart of FIG.

In FIG. 9, the motion vector evaluation circuit 12
First, in step P1, the block-by-block motion vector detected by the motion vector detection circuit 5 is acquired.
Furthermore, the motion vector evaluation circuit 12 acquires the minimum correlation value in block units detected by the correlation value detection circuit 11 in step P2 and the correlation values around it.

Then, in step P3, the motion vector evaluation circuit 12 calculates the correlation value gradient between the minimum correlation value fetched as described above and the surrounding correlation values for a certain evaluation block. That is, in the example of FIG. 10, the position c0 of the minimum correlation value and the position c of the surrounding correlation values
The correlation value gradient in eight directions between 1 and c8 is calculated.

It should be noted that the calculation of the correlation value gradients in the eight directions in this way is merely an example, and for example, one of the minimum correlation value position c0 and the surrounding correlation value positions c1 to c8 can be used. You may make it calculate the correlation value gradient of four directions between four positions.

Then, in step P4, the motion vector evaluation circuit 12 calculates the maximum value, the minimum value, and the average value of the plurality of correlation value gradients obtained in step P3. Then, in step P5, the motion vector evaluation circuit 12 determines whether or not the difference between the maximum value and the minimum value of the correlation value gradient obtained in step P4 is larger than a predetermined threshold TH1 or whether the correlation value gradient It is determined whether or not the average value is smaller than a predetermined threshold TH2.

If any one of these two conditions is satisfied, the process proceeds to step P6 to invalidate the motion vector of the corresponding evaluation block. On the other hand, when neither of the above two conditions is satisfied, step P
The process of step 6 is skipped and the process proceeds to step P7.

Next, in step P7, it is checked whether or not the above-described motion vector evaluation has been completed for all blocks. If there is an unprocessed block, step P
Returning to step 3, the processes of steps P3 to P7 are repeated until the motion vector is evaluated for all the blocks.

When the motion vectors have been evaluated for all the blocks, the process proceeds from step P7 to step P8, and only the valid motion vector s16 which has not been invalidated by the above processing is determined by the motion vector determination circuit 6
Supply to. The motion vector determination circuit 6 determines the motion vector s17 of the entire field using only this valid motion vector s16.

As described above, in this embodiment, the motion vector of each block is evaluated based on the predetermined correlation value gradient, and the motion vector of the entire field is determined by using only the motion vector validated as a result of this evaluation. By doing so, the detection accuracy of the motion vector can be improved. Here, the reason why the detection accuracy can be improved in this way will be described with reference to FIG. 11.

FIG. 11 is a characteristic diagram showing the state of the distribution of the correlation value according to the image situation and the state of the change of the correlation value gradient near the minimum correlation value. Note that c1 to c8 in FIG.
Corresponds to the positions c1 to c8 of the surrounding correlation values shown in FIG.

The correlation value distribution in the case where the condition of the target image is good and the correlation is easily obtained is shown in FIG.
As shown in. It can be said that such a correlation value distribution is an ideal correlation value distribution in which it is easy to specify the minimum correlation value.

On the other hand, the spatial frequency of the image is low,
The correlation value distribution when there are few characteristic patterns is as shown in FIG. Further, the correlation value distribution when there is a strong correlation in the specific direction regardless of the movement of the subject is as shown in (c) of FIG. 11. It can be said that these correlation value distributions are correlation value distributions in which it is difficult to specify the minimum correlation value and the motion vector is likely to include an error.

Here, paying attention to the correlation value gradient near the minimum correlation value, the correlation value gradient in the correlation value distribution as shown in FIG. 11A is as shown in FIG. The gradient is uniform in the direction and has a value equal to or greater than a certain value.

On the other hand, the correlation value gradient in the correlation value distribution as shown in FIG. 11B is uniform in both directions as shown in FIG. 11B ′, Its absolute value is extremely small. Also, FIG. 11 (c)
In the correlation value gradient in such a correlation value distribution, the gradient in a specific direction is significantly different from the gradients in other directions, as shown in (c ′) of FIG.

Therefore, in the correlation value distribution as shown in FIG. 11B, the average value of the correlation value gradient is the average value of the correlation value gradient in the correlation value distribution as shown in FIG. 11A. It will be small in comparison. Also, FIG. 11 (c)
In the correlation value distribution as shown in Fig. 11, the difference between the maximum value and the minimum value of the correlation value gradient is the difference between the maximum value and the minimum value of the correlation value gradient in the correlation value distribution as shown in Fig. 11A. It will be a big thing in comparison.

Therefore, as in the present embodiment, the difference between the maximum value and the minimum value of the correlation value gradient and the average value of the correlation value gradient are compared with appropriate predetermined values, respectively, to thereby obtain (b) in FIG.
It is possible to discriminate between the correlation value distribution as shown in FIG. 11C in which an error is likely to occur in the motion vector and the optimum correlation value distribution as shown in FIG.

The motion vector detection accuracy can be improved by detecting the motion vector of the entire field using only the motion vector having the optimum correlation value distribution determined in this way.

In the third embodiment, the maximum / minimum value and average value of the correlation value gradient are used to determine whether the motion vector is valid or invalid. It is not limited to.

For example, if DEFRMS representing the strength of the correlation value distribution with respect to the disturbance in the spatial direction as expressed by the following (7) is equal to or greater than a predetermined value, or DEFAVE representing the average value of the correlation value gradient is Even if the motion vector of the corresponding block is invalidated when it is equal to or less than the predetermined value, the same effect as described above can be obtained.

DEFRMS = Σ (DEFAVE−CORDEF (i)) ² / (i−1) (Equation 7) Here, CORDEF (i) represents individual correlation value gradients in a plurality of directions, and i is a correlation value. The number of gradients is shown.

FIG. 12 is a flow chart showing the procedure of motion vector evaluation in such a case.
Only the processing contents of steps P4 and P5 differ from the flowchart shown in FIG.

That is, in the example of FIG. 12, in step P4, the motion vector evaluation circuit 12 uses the correlation value gradients in a plurality of directions obtained in step P3 to determine the gradient disturbance strength DEFR.
Calculate MS and mean DEFAVE.

Next, in step P5, the motion vector evaluation circuit 12 determines the gradient turbulence DE determined in step P4.
It is determined whether FRMS is larger than a predetermined threshold TH1 or whether the average value DEFAVE of the slope is smaller than a predetermined threshold TH2.

If any one of these two conditions is satisfied, the process proceeds to step P6 to invalidate the motion vector of the corresponding evaluation block. On the other hand, when neither of the above two conditions is satisfied, step P
The process of step 6 is skipped and the process proceeds to step P7. Then, the processes of steps P3 to P7 are repeated until such a motion vector is evaluated for all blocks.

Next, when the motion vectors have been evaluated for all the blocks, the process proceeds from step P7 to step P8, and only the valid motion vectors that have not been invalidated by the above processing are passed to the motion vector determination circuit 6. Supply. The motion vector determination circuit 6 determines the motion vector of the entire field using only this valid motion vector.

FIG. 13 is a flow chart showing the main operation of the motion vector detecting device according to the third embodiment. In FIG. 13, first, in step P1, an image signal to be a motion vector detection target is binarized at a zero level. Then, in step P2, with respect to the binarized image signal, the correlation calculation between the image signal of the current field and the image signal of the previous field is performed in block units according to the block matching method.

Then, in step P3, this step P
A motion vector in block units is detected based on the correlation value in block units obtained by the correlation calculation of 2. Specifically, the block in the previous field having the minimum correlation value is searched for, and the relative shift of this block is detected as a motion vector in block units.

In step P4, the minimum value of the block-by-block correlation value obtained by the correlation calculation in step P2 described above and the correlation values of the surrounding positions separated by a predetermined pixel from the position where this minimum correlation value is present. And are detected.

Further, in step P5, based on the minimum correlation value thus detected and the correlation values around it,
The same processing as the processing shown in the flowchart of FIG. 9 or the flowchart of FIG. 12 is performed. As a result, it is determined whether the motion vector of each block is valid or invalid.

Then, in step P6, only the motion vector determined to be a valid motion vector in step P5 is used to determine the motion vector of the entire field. Specifically, the median value or the average value of the target block-based motion vectors is determined as the motion vector of the entire field.

[0102]

As described above, according to the present invention, the reliability of the motion vector detected by the motion vector detecting means is determined, and based on the result of this determination, the specific motion vector to be detected is determined. Since it is determined, the block-based motion vector that causes an error in the specific motion vector is invalid, and only other useful block-based motion vectors are used to determine the specific motion vector. Therefore, the accuracy of detecting the motion vector can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a shake prevention device to which a motion vector detection device according to a first embodiment is applied.

FIG. 2 is a diagram for explaining how a block-by-block motion vector is detected from an image input to the motion vector detecting device according to the first embodiment, and classification information is generated based on the block-by-block motion vector. It is a figure.

FIG. 3 is a flowchart showing the operation contents of the motion vector classification circuit according to the first example.

FIG. 4 is a block diagram showing a configuration of a shake prevention device to which the motion vector detection device according to the second embodiment is applied.

FIG. 5 is a diagram for explaining how a block-by-block motion vector is detected from an image input to the motion vector detection device according to the second embodiment, and classification information is generated based on this block-by-block motion vector. It is a figure.

FIG. 6 is a flowchart showing the operation contents of the motion vector classification circuit according to the second embodiment.

FIG. 7 is a flowchart showing main operation contents of the motion vector detection device according to the first embodiment or the second embodiment.

FIG. 8 is a block diagram showing a configuration of a shake prevention device to which the motion vector detection device according to the third embodiment is applied.

FIG. 9 is a flowchart showing the operation contents of the motion vector evaluation circuit according to the third embodiment.

FIG. 10 is a diagram showing a spatial arrangement of correlation values near the minimum correlation value.

FIG. 11 is a characteristic diagram showing a correlation value distribution according to an image situation and a change state of a correlation value gradient near the minimum correlation value.

FIG. 12 is a flowchart showing another operation content of the motion vector evaluation circuit according to the third embodiment.

FIG. 13 is a flowchart showing main operation contents of the motion vector detection device according to the third embodiment.

FIG. 14 is a block diagram showing a configuration of a conventional shake prevention device.

[Explanation of symbols]

 1 memory 2 spatial frequency filter 3 correlation calculation circuit 4 memory 5 motion vector detection circuit 6 motion vector determination circuit 7 memory read control circuit 8 motion vector classification circuit 9 motion vector classification circuit 10 binarization circuit 11 correlation value detection circuit 12 motion Vector evaluation circuit

Claims (4)

[Claims] 1. A correlation calculation means for performing a correlation calculation between images continuous in time series, a motion vector detection means for detecting a motion vector based on the correlation value calculated by the correlation calculation means, and the motion. Based on the motion vector detected by the vector detection means, a motion vector determination means for determining a specific motion vector, and a motion vector and another motion vector for each motion vector detected by the motion vector detection means. And a motion vector classifying unit that classifies the motion vectors based on a difference between the two motion vectors obtained as a result of the comparison and derives the classification result to the motion vector determining unit. Motion vector detection device. 2. The correlation calculation and motion vector detection,
The motion vector detecting apparatus according to claim 1, wherein the motion vector classification is performed in a predetermined block unit. 3. The motion vector determination means determines a specific motion vector by using a block-unit motion vector of a valid classification among the block-unit motion vectors classified by the motion vector classification means. The motion vector detecting device according to claim 2, wherein 4. A correlation calculation between time-sequential binary images composed of a binarizing unit for binarizing an image signal and an image signal binarized by the binarizing unit. Correlation calculation means, motion vector detection means for detecting a motion vector based on the correlation value calculated by the correlation calculation circuit, and a specific motion vector determined based on the motion vector detected by the motion vector detection means Motion vector determining means, correlation value detecting means for detecting the spatial distribution of the correlation values calculated by the correlation calculating means, and change in the correlation value gradient in the spatial distribution of the correlation values detected by the correlation value detecting means. Based on the absolute value, the reliability of the motion vector detected by the motion vector detecting means is evaluated, and the evaluation result is derived to the motion vector determining means. Motion vector detecting apparatus characterized by comprising a vector evaluation unit.