JPH0435384A - Moving vector detector - Google Patents

Abstract

PURPOSE:To prevent an erroneous moving vector from being detected by providing a moving vector detection section, a correlation sort circuit, a moving vector discrimination section, a moving vector - distance conversion circuit, a sort data discrimination circuit and a moving vector discrimination section to the detector. CONSTITUTION:Correlation values sorted in the higher order and their relevant deviations are obtained by a correlation sort circuit 13. The deviation data is converted into a distance from the deviation with a highest deviation by a deviation - distance converter 14, and the correlation is normalized by the deviation at the highest correlation at a correlation normalized circuit 16. A sort data discrimination section A15 discriminates the presence of the data by using the correlation and the distance and outputs valid/invalid information 1. A correlation validity discrimination circuit 9 discriminates whether or not the moving vector is valid depending on the correlation and outputs valid information 2. The discrimination circuit 10 collates the valid information sets 1,2 to decide the entire moving vector from the moving vector corresponding to the split area discriminated to be valid.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a motion vector detection device for detecting the amount of motion of an image.

2. Description of the Related Art An example of a conventional image motion vector detection device is the one shown in Japanese Unexamined Patent Application Publication No. 61-269475.
1 is latch A, 2 is representative point memory, 3 is latch &
4 is a correlator, 5 is an address controller, 6 is an address switching section, 7 is an accumulation adder, and 8 is a correlation search section.
Reference numeral 9 indicates correlation validity/invalidity determination. 10 indicates a determination circuit. An image motion vector detection device using a conventional correlation calculation device configured as described above will be explained. First, image motion vectors will be explained. Also Figure 14 (
a)i;L Even if it shows an image at a certain time, and (b) shows an image one field or one frame later, when the image moves in parallel due to the movement of the imaging device, etc., (C) As shown by the arrow, a vector representing the amount by which the image is translated in parallel is called a motion vector.A method for detecting a motion vector will be explained.

Figure 15 shows representative points (an example of nine representative points) and the surrounding pixels in the representative point matching method, which is the most common method for detecting motion vectors in images. First, motion vector detection is 1.
In each divided area, the screen is divided into multiple areas, and in each divided area, correlation calculation is performed to determine where the image data at the position of the representative point in a certain field has moved among the surrounding pixels in the next field. A motion vector is detected by statistically calculating the result for all representative points.Then, based on the multiple motion vectors detected in each divided area, a motion vector representing one screen is finally determined. The motion vector detection method using correlation calculation will be explained in more detail with reference to FIGS. 15(a) and (b). ) at the representative point 511-+
Calculate the absolute value of the difference between the image data at the position (Xk, Yb) and each pixel data in the area around the representative point (motion vector detection area) at the next time n (field or frame). This calculation is performed for each representative point and the area around it. Then, when each representative point is used as a reference point, the absolute difference calculation results for locations with the same relative position are cumulatively added. In this case, the number of representative points is 9, so the cumulative addition is based on 9 pieces of data. Addition This cumulative addition produces the same number of numerical data as the number of pixels included in the motion vector detection area, that is, the correlation value P(i).

Even if j) is obtained The above operations can be generally expressed as follows.

P (i, j) = Σ l S, I-+ (Xk, Yh) Sn
(X k-+, Yk-+) l
... (1) However, 5-(Xk, Yb) is the coordinate (Xk.

P(i, j) is the correlation value at the relative coordinates (i, j) with respect to the representative point, and Σ is the cumulative addition for the range of the subscript, that is, the number of representative points. Indicates that something is done.

In this case, k is a value from 1 to 9.
The range of change in 1 r J represents the motion vector detection range. In addition, in the calculation of the first equation, the correlation calculation is performed by shifting the image by i and J, so from now on, L J will be referred to as a shift. The smaller the value, the higher the correlation. In this case, the number treats two images obtained by parallel movement as input signals. There is only one state in which the two images exactly overlap,
The deviation (i, j) indicating this state represents a motion vector. In order to obtain a, that is, the motion vector V(i, j), the following equation is calculated.

V(i, j) = (i, j) for m1n(P(i,
j))...(2) However, Lmin is an operator that selects the one with the minimum value. By the above calculation, the motion vector V
The method for determining (i, j) is called the representative point matching method, and in this way, the motion vector in each divided area can be detected even if it is suboptimal. The image data at each representative point on the screen is taken into the latch A1 by the timing pulse LPI, and at the appropriate timing, the image data at each representative point on the screen is transferred to the address corresponding to each representative point in the representative point storage memory 2. Then, in the next field or next frame, the image data in the motion vector detection area around the position of each representative point is correlated with the representative point of the previous field stored in the representative point memory 2, and the cumulative Input to the adder 7 The cumulative adder 7 cumulatively adds the correlated data at the same coordinate position with respect to the representative point, and then the cumulative addition around all the representative points is completed. Then, the correlation search circuit 8 determines the location with the highest correlation value among the cumulative addition values held in the cumulative adder 7. In other words, when the position of the representative point is used as a reference, the location with the highest correlation value is determined. The position (address) having a high value is the most likely to be a motion vector. Based on the distribution of correlation values in the motion vector detection area, the correlation validity/invalidity determination circuit 9 determines that the motion vector obtained by the correlation calculation is Determining whether it is valid or invalid Specifically, the determination is made using the average value, minimum value, and maximum value. Joint plate If the following conditions are not met, the motion vector detected from that detection area is invalid and removed from the candidates for motion vector determination.

Average value of correlation values>A...(3) Average value of correlation values>Minimum value of BX correlation values...(4) Minimum value of correlation values/Maximum value of correlation values<C...(5) (However, B and C are predetermined constants.) Motion vector detection and correlation validity/invalidity determination using the representative point matching method described above is performed as shown in FIG. 15(a).
Even if multiple motion vectors are determined to be valid in one screen by being detected at multiple locations on the screen, the motion vectors determined to be valid are determined to be valid based on the motion vectors obtained from each divided area of the screen and their validity determination information. As a determination method, the motion vector of the entire screen representing one screen is determined using a plurality of motion vectors. Although the method of using the entire motion vector is often used, the above operation is performed for each field (frame).While performing the correlation calculation, it is necessary to save the image data at the representative point for the correlation calculation of the next field (frame). There is a latch A1.Also, a latch B 341 holds the image data of a representative point when correlating the image data of a certain representative point with the surrounding image data. A motion vector detection unit 11 is used to detect motion vectors, and a motion vector determination unit 12 is used to determine the motion vector of the entire screen from the motion vectors of each area of the screen obtained by the motion vector detection unit and their correlation information. Problems to be Solved by the Invention However, in the above configuration (if the friendship image itself has a strong correlation, for example, if the image has only a striped pattern of vertical lines or horizontal lines), the distribution of correlation values after correlation calculation is ( In such a case, the correlation search circuit 8 determines only one point with the highest correlation, and if that point is used as a motion vector, the position with the highest correlation can be found. There is a problem in that the motion vector changes greatly due to noise in the image, causing a large error in the output motion vector.In order to solve this problem, the present invention determines when there is a strong correlation in the input image itself, and detects incorrect motion. A motion vector detection unit, a correlation value sorting circuit, a motion vector determination unit, a motion vector-distance conversion circuit, a sorted data determination circuit, The present invention provides a motion vector detection device characterized by having a motion vector determination section.The present invention uses the above-described structure to sort correlation values calculated in motion vector detection for each detection area, and to sort data with a large correlation. When the state of the input image is identified using data up to a certain predetermined rank, and it is predicted that an incorrect motion vector will be detected, the motion vector of the entire screen is determined without using the motion vector of that detection area. Embodiment 1 FIG. 1 is a block diagram of a motion vector detection device according to a first embodiment of the present invention, and 11 is a motion vector detection device ff.
1L 12 is a motion vector determination unit, which has the same configuration as the conventional example.It differs from the conventional example (correlation value sorting circuit 13, deviation-distance converter 14, and sort data determination unit A).
15 and a correlation value standardization circuit 16 are inserted, and the motion vector determination unit 12 also uses the output of the sort data determination circuit 15.
The operation of the motion vector detection device configured as above will be explained below because it outputs the motion vector of the entire screen as the final output. The motion vector in each divided area is expressed as (1) (2
) and outputs the motion vector and the distribution of correlation values in the motion vector detection area.The distribution of this correlation value will be explained in detail here.Figure 2 shows the correlation value P(i, j) in the motion vector detection area. is plotted against one-dimensional deviation l or j.
As shown in Figure (a), the correlation value shows sharp changes, and the shift with the highest correlation (in this case, the shift that takes the minimum value) represents the motion vector. Figure 2(b) shows a two-dimensional representation of the distribution of correlation values in this case.In this way, parts with high correlation within the motion vector detection area appear only in the very narrow l area, and that part is subject to movement. Represents a vector.

In contrast to this situation, in the case of a subject that has a high correlation with the input image itself, for example, when a diagonal bar-shaped object is photographed against a patternless background as shown in Figure 3 (a), Because of the correlation between If it is a motion vector, there is a possibility that any part of l'& in the area shown in black in Figure 3(b) will be determined to have the highest correlation, and there is a possibility that the output motion vector will fluctuate greatly. - When an image with only vertical lines is input as shown in Figure 4(a), correlated shifts appear periodically, and points with high correlation appear as shown in Figure 4(b). Multiple occurrences
In this case as well, the motion vector that is output varies greatly.

In this way, the purpose of the present invention is to determine when the correlation of the input image itself is large and to control the motion vector detection device so as not to output an erroneous motion vector. The numbers are given as examples of the points of shift at which they appear in the motion vector detection area. That is, the shift (i, j) indicated by number 1 is determined to be the motion vector. Figure 5(a)
is the case of a normal object with high contrast, and even if the points with high correlation values are clustered near the shift with the highest correlation, in contrast, in Fig. 5(b), there are bar-shaped objects on a patternless background. Even in the case of an image with an object, in this case, the upper correlation value data is located at the diagonal position of the shift with high correlation. The upper correlation value data is arranged in multiple lines.The upper correlation value data (correlation value/distance from the point with the most correlated deviation) in these states (a) to (c) is graphed. Examples of this are shown in Figures 6 (a) to (c). In the case of the state f(a), that is, a normal high-contrast subject, when the area is first, the state of (b), that is, a patternless background. In the case of an image with a rod-shaped object, the shape of the region where data appears is like region 3, and some data also appears in region 2 (B in the figure).In other words, the correlation value increases as the distance increases. (Even if parts with small correlation values sometimes occur, these prominent parts with small correlation values ゛How many forces appear in B\\ Although the part in which they occur varies depending on the human image, areas like this occur. This is because the highly correlated parts exist at points other than point A, which indicates the motion vector.
It is similar to b), and the data appears in area 4, and data also appears in area 2.A To summarize the above, the distance between states (b) and (c) and state (a) is Are there any points where the correlation value decreased when it was large (
In other words, when data exists in region 2 shown in Figure 6 (a), it can be determined from the distribution of correlation values. It is possible to prevent incorrect motion vectors from being detected by disabling the motion vectors and determining the motion vector for the entire screen from only the valid motion vectors. We measure the variation in the range (standard deviation of correlation values at each distance) and determine region 2 by taking into account the variation.We will use the configuration diagram in Figure 1 to explain how the operation described above is realized. The motion vector detection unit 11 may obtain motion vectors and data on the distribution of correlation values in the motion vector detection area (correlation values and their corresponding deviations). The data is input to the sorting circuit 13 and is sorted in order from the most correlated data to a predetermined rank. Here, data with a high sorting rank, that is, data with a large correlation (those with a small P (i, j)) are prioritized, and those with a low sorting rank are In other words, those with low correlations (those with large P(i, j)) are classified as low-ranking items.This allows data on correlation values and their corresponding deviations to be obtained, sorted from top to bottom in descending order of correlation. The deviation data of this plate is converted into distance from the deviation considered to be most correlated (the deviation ranked No. 1 in the sorting process) by the deviation-distance converter 14. An example is shown in the following equation.

Dn=Iin-itl+1jn-j+l-(6)
However, [)n is the distance fltL (it, j+
) is the most correlated deviation (indicating the motion vector), (
t'+j++) is the shift determined to have the nth highest correlation in the sorting process.A Another way to calculate distance is ζ.3
There are methods for calculating distances in two-dimensional data using the angle method.
It is converted to a value based on the smallest correlation value among the correlation values that have become the number. An example of the conversion method is shown in the following equation.

P' (in, jn) = P (iR, jn) P (it, j+) ... (
7) However, Po is the correlation value after conversion.The sorted data determination unit A15 GEL uses the correlation value and distance that have been subjected to the sorting process obtained as above.
It is determined whether or not data exists in area 2 in Figure 6(a) within a certain predetermined distance, and if data exists in this area, the motion vector obtained in that divided area is invalidated. Invalidity information 1 is output to the judgment circuit 10. The range of sorting ranks to be checked here varies depending on the conditions of the input image.
It has been found through experiments that if the correlation value is less than this, it is difficult to distinguish between a normal image and an image with a high correlation value.

Also, similar to the determination in the conventional example, the correlation validity/invalidity determination circuit 9 determines whether the motion vector obtained by the correlation calculation is valid or invalid. The judgment conditions are equations (3) to (5), and if any of these conditions is not satisfied, the motion vector detected in that divided area is invalidated, and the validity/invalidity information 2 is input to the judgment circuit 10. Then, the determination circuit 10 (table 1) refers to the valid/invalid information 1 and 2 and determines the motion vector of the entire screen from only the motion vectors corresponding to the divided areas that are both determined to be valid.
As a determination method, a method is used in which the average value or intermediate value of a plurality of motion vectors determined to be valid is used to determine the motion vector for the entire screen. 7 is a block diagram of a motion vector detection device according to a second embodiment of the present invention. 11 is the number of motion vector detections, 12 is the motion vector judgment lL, and 13 is the number of correlation value sorting times j! &, 17 is a sort data judgment unit B, which differs from the first embodiment in that it judges the state of correlation values. The operation of the motion vector detection device configured as described above will be explained below.The motion vector detection unit uses the correlation value data that is The motion vector in each divided area is calculated as described above (
1) Calculate using equation (2) and output the motion vector and the distribution of correlation values in the motion vector detection area.a Here, in the states shown in Figures 5(a) to (C), how do the sorted correlation values look like? 8(a) to (C) show the relationship between the sorting order and the correlation value corresponding to the state of each input image in FIGS. 5(a) to (c). In this way, when the contrast is sufficient in a normal input image, that is, in case (a), the correlation value increases rapidly according to the rank (curve A in the figure). On the other hand, in the case of (b) and (C), even if the magnitude of the correlation value remains small at a certain lower rank, since the correlation of the image itself is high, the part with the small correlation value is (a ) This tendency also occurs when the contrast of the input image decreases and the motion vector detection accuracy decreases. Area 5 shown in (a)
If there is data in the area, it can be determined that the motion vector has not been detected correctly.In other words, the motion vectors of the divided area where such a situation occurs are invalidated, and only the motion vectors that are valid are invalidated. If the motion vector of the entire screen is determined from The motion vector detection unit 11 generates motion vectors and data on the distribution of correlation values in the motion vector detection area (
This correlation value distribution data is input to the correlation value sorting circuit 13 and is sorted in order from the most correlated data to a predetermined rank. The data of the correlation values sorted in order and the corresponding deviations are obtained, and the sorted correlation values are processed by the sort data determination unit B17.
In each divided area, check whether the correlation value in the sorting order within a predetermined range is included in area 5 shown in (a) of FIG. 4. Setting of area 5 in Fig. 8(a) (like
Plot the data of the correlation value in the case of an input image with normal contrast, consider the standard deviation of that data, and assume that the values below that are in region 5 (~As described above, the 8th If correlation value data exists in the shaded area in Figure (a), the motion vector of that divided area is invalidated, and valid/invalid information 1 is output to the determination circuit 10, which is also the same as the determination in the conventional example and the first embodiment. The correlation validity/invalidity determination circuit 9 determines whether the motion vector obtained by the correlation calculation is valid or invalid.The determination condition is expressed by equation (3).
- (5), and if any of these conditions are not met, the motion vector detected in that divided area is invalidated and valid/invalid information 2 is input to the determination circuit 10. Then, the determination circuit 1041 Valid/invalid information According to the present embodiment, as described above, according to the present embodiment, the correlation values after the sorting process are determined. By simply using this function, it is possible to judge both when the input image itself has a strong correlation and when the contrast of the input image has decreased, and to prevent the detection of erroneous motion vectors. Figure 9 (a) shows the change in the cumulative correlation value in this case. ~(c)
However, in the case of cumulative correlation values (as in the graph of correlation values in Figure 8), the characteristics can be obtained by accumulating the correlation values. In the case of a), the slope is large (curve B in the figure), and the correlation between the other images themselves is high (b), and in (c), the slope is small.
Taking advantage of this, the correlation value determination unit B17
), check whether correlation value data is included in region 6 (~Here, region 6 is determined by measuring boundary line B and taking into account the standard deviation of that data, but Figure 10 is This is a block diagram of a motion vector detection device according to a third embodiment of the present invention, in which 11 is a motion vector detection section, 12 is a motion vector determination section, 13 is a correlation value sorting time K, and 18 is a shift-distance dispersion conversion. Section 19 is a sort data judgment section C, which differs from the first embodiment in that it judges the state of the correlation value.In addition to the judgment of the conventional example,
The operation of the motion vector detection device configured as described above is explained below, using the variance of the distance from the most correlated deviation sorted by the magnitude of the correlation value up to a certain predetermined rank. First, in the same manner as in the conventional example, first and second embodiments, the motion vector detection section calculates the motion vector in each divided area of the screen using the above-mentioned equations (1) and (2), and detects the motion vector and the motion vector. Outputting the distribution of correlation values in the region Now we will explain how the sorted correlation values change in the states shown in Figures 5 (a) to (C).
Figures 1 (a) to (c) show the relationship between the sorting rank corresponding to the state of each input image in Figures 5 (a) to (c) and the variance of the distance from the most correlated shift to that rank. Even if the contrast is sufficient in a normal input image like this, in case (a), the dispersion of distance increases according to the order (curve C in the figure).In contrast, when the input image itself In cases (b) and (c) where the correlation is high, the variance of the distance increases rapidly and tends to saturate.

This is because the correlation of the images themselves is high, and the parts with low correlation values exist in a wider range than in a normal input image like (a). Area 7 shown in a)
If there is data in the area, it can be determined that the motion vector has not been detected normally.In other words, the motion vector of the divided area where this condition occurs is invalidated.
By determining the motion vector of the entire screen from only the valid motion vectors, it is possible to prevent erroneous motion vectors from being detected. First, the motion vector detection unit 11 generates data (
This correlation value distribution data is input to the correlation value sorting circuit 13 and is sorted in order from the most correlated data to a predetermined rank. The data of the correlation values sorted in order and the data of the corresponding deviations are obtained. (The distance is the same as in the first example).The formal method is to calculate the square of the difference between the average value and each data. There is no problem in this embodiment in which the extent of the L data is taken into account by taking the absolute value of None (°C) Expressing this operation as a formula,
As follows, σn= (E (ave Dr+ −Dn) ”)
/N...(8) However, σ. is the variance a of the distance D11 to the nth sorting rank
ve I)+ is the distance to the nth sorting order, and the average KN is the total number of data. The variance data ζ of this distance is not directly related to the contrast of the input image, so it is not directly related to the contrast change of the input image. Although it is possible to judge the state of the correlation value without being influenced too much, input the variance of the distance to each sorting order into the sort data judgment unit C19a Sort data judgment unit C19ζ Check whether the variance of the distance to the sort order in the range of is included in the area 7 shown in Fig. 11(a). Plot the data of the dispersion of distance in the case of an input image with a contrast of .Considering the standard deviation of the data, let us assume that the values above this are in the region 7 (~ As shown in Figure 11). If correlation value data exists in region 7 of (a), the motion vector of that divided region is invalidated, and validity/invalidity information 1 is output to the determination circuit 10. The circuit 9 judges whether the motion vector obtained by the correlation calculation is valid or invalid.The judgment conditions are equations (3) to (5), and if any of these conditions is not satisfied, the divided area is Invalidate the motion vector detected in
The valid/invalid information 2 is input to the determination circuit 10, and the valid/invalid information 1 and 2 are referred to, and the motion vectors of the entire screen are selected from among only the motion vectors corresponding to the divided areas that are both determined to be valid. As described above, according to this embodiment, when the input images themselves have a strong correlation, it is determined based on the dispersion of the distance at the top of the sorted data. &ζ This can prevent the detection of erroneous motion vectors, and also the sorted correlation values or cumulative correlation values and distances used to determine the state of the correlation values in the second and third embodiments. The variance of This shows the area that appears when the processed cumulative correlation value and distance variance are plotted as two-dimensional data. Area 9 in Figure 12(a) is a normal image, that is, when there is no strong correlation in the input image itself. Here, the boundary line of regions 8 and 9 is defined as D. On the other hand, region 10 in (b) and (c) is a region where the input image itself has a strong correlation. In this way, the characteristics shown in FIGS. 9 and 11 in Examples 2 and 3 described above may appear.
Therefore, if the cumulative correlation value is low, the motion vectors in the divided areas containing data in areas with large distance dispersion, that is, area 8 in Figure 12(a), should be determined to be invalid (~ here In this case, area 8 can be determined by measuring the boundary line and taking into account the standard deviation of the data. Good judgment can be made.Also, in this case, the force using the cumulative correlation value (the same judgment can be made using the correlation value itself).In the first embodiment, the sort data judgment unit A15J! There is no need to check the data for all sort orders (°C or In the second embodiment, a correlation value standardization circuit 16 is inserted between the sumo wrestler correlation value sorting circuit 13 and the sorted data judgment unit B17 using the sorted correlation values, and the correlation value is based on the minimum value of the correlation values. You can also convert it to a value and use it (
In addition, in the first to third embodiments, if all the signal processing is digitalized, it is easy to realize all or part of these processes on microcomputer software, and the effects of the present invention According to the paper, it is possible to construct a motion vector detection device that can determine when the input image itself has a strong correlation and prevent the detection of erroneous motion vectors, and its practical effects are significant (~

[Brief explanation of the drawing]

Fig. 1 shows the configuration of a motion vector detection device according to the first embodiment of the present invention. Fig. 2 shows the one-dimensional relationship between the correlation value and the deviation. Fig. 3 shows the configuration of the motion vector detection device according to the first embodiment of the present invention. Figure 4 shows the input image and correlation values when an image with only vertical lines is input.
Fig. 5 shows the relationship between deviation and sort order. Fig. 6 shows the relationship between distance and correlation value. Fig. 7 shows the second example of the present invention.
FIG. 8 shows the block diagram showing the relationship between the sort processing order and the correlation value. FIG. 9 shows the relation between the sort processing order and the cumulative correlation value. Block showing the configuration of the third embodiment. Figures 1 and 3 show the configuration of a conventional motion vector detection device. Figure 14 explains the concept of motion vectors. Figure 15 explains the relationship between representative points and pixels on the screen and surrounding areas in the representative point matching method. In the diagram, 1...Latch A, 2...Representative point memory, 3.
...Latch & 4...Correlator 5...Address controller, 6...Address switching same section 7...
Accumulative adder 8...Correlation search section 9...Correlation validity/invalidity judgment section 10...Judgment section 11...Motion vector detection block 12...Motion vector judgment section 13
. . . Correlation value sorting circuit 14 .
・Shift-distance dispersion conversion eye 19... Name of agent for sorting data judgment Patent attorney Shigetaka Awano Figure 1 (Yun correlation g L special (f > Z original) 3 g (b Jrr cloth (2 originals) Fig.
) 槓 Closed ihensuri minute map p Figure input transition i act (dedicated 1 person's academic degree ton (bn3g interval fu 1 minute 1 figure (t > at Gositrus F's high input passage 111 ( L
JftLI (HLtt f, shir-4?l) Fig. ta->ff1lt's hun) Raz)'s special-entering you 1 shi elephant tb) Feather 1 large object book 3; force image Fig. hi) iiLt's low contrast input image iiCb) Rod-shaped #'l! Elephant sort to Yakasho #Tribute number diagram S hector ramie! Ramie diagram (L) M Illustrated illustration of the city's trust (b)
@ Tato's 8 and 5 ME (C) Vertical line only Heka direct Izo) 11 Stomach Iyu B Figure 12 (α) Normal contrast island yV input 5 lines ((ν) Haku I Input with #!J in input 111 t(,) l Line only kuiriba 11 Ikame No. 1 Figure 1 (iL) AI's high input image of Fucitrus F Ikame sorting #l center L (b) $ Sort 111 of the image of melted rice
Responsibility LL (Q) Entrance to the school only for the Kabuki boat school) II I1γ 2nd section

Claims (4)

[Claims] (1) A motion vector detection unit that divides the input image signal into multiple regions and calculates the motion vector of the image in each region, and determines the motion vector of the entire screen by determining the effectiveness of the motion vector in each region. The motion vector detection device includes a correlation value sorting unit that sorts correlation values and shift information indicating positions corresponding to the correlation values according to the magnitude of the correlation values; 1. A motion vector detection device comprising a sorted data determination unit that determines the validity of a motion vector in each region using at least correlation values up to the ranking and shift information corresponding to the correlation values. (2) The sorted data determination unit determines the validity of the motion vector in each region using the correlation values or cumulative correlation values from the third predetermined rank to the fourth predetermined rank that have been sorted. The motion vector detection device according to claim 1. (3) The sort data determination unit uses the variance of the distance from the position with the highest correlation, that is, the shift, to the rank shift in a fifth predetermined range, obtained by the sorting process, to determine the movement in each area. The motion vector detection device according to claim 1, wherein the motion vector detection device determines the validity of the vector. (4) The sort data determination unit determines the distance from the position with the highest correlation, that is, the shift, to the rank shift in a sixth predetermined range, and the correlation value or Claim 1 characterized in that the validity of the motion vector in each region is determined using the cumulative correlation value.
The motion vector detection device described in .