JP2003264757A - Retrieval device and retrieval program for video image - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide the retrieval device and retrieval program of video images for efficiently detecting a human body such as an announcer in input video images. <P>SOLUTION: A video input part 12 successively fetches out video data including news video images or the like from a storage device 11 and inputs them to a segment detection part 13 by control from a control part 10. The segment detection part 13 detects the change of the scene of the input video images. The time and video data of a cut point detected in the segment detection part 13 are inputted to a motion extraction part 17 and the motion information of each segment is extracted. In a specified color distribution detection part 18, a specified color distribution inside the segment is detected. A human body image determination part 15 determines the segment where motion inside the segment is small and the specified color distribution is present inside the segment as the segment where a human body image is present. Also, a segment length detection part, a telop detection part and a small area color change measurement part each can be added. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a video search device, and more particularly to a video search device and a search program capable of efficiently detecting a person such as an announcer from video data or moving image data.

[0002]

2. Description of the Related Art A conventional example of automatically detecting an announcer screen from a news video will be described with reference to FIG. In the figure, the video input section 12 is controlled by the control section 10 to sequentially take out video data such as a news video from the storage device 11 and input it to the segment detection section 13. The segment detection unit 13 detects the switching of the scene of the input video, and outputs the screen data at the switching point and the time thereof to the segment correlation unit 14.

The segment correlation unit 14 measures the correlation value between the segment head screens. As a method of measuring the correlation value, there is a method of measuring the correlation value of the color histogram between the screens, and the invention "a similar image search device and method" described in Japanese Patent Laid-Open No. 08-249467 is used. You can In this method, as shown in FIG. 12, using the first segment as a key image, the correlation between the segment leading screens between the first segment and the second to nth segments is measured, and then the second segment is measured. Correlation measurement between the 3rd to nth segment is performed, and hereinafter,
~ Correlation measurement between the nth segment, ..., Correlation measurement of the (n-1) th segment and the nth segment is performed. And
The number of segments whose correlation value is a predetermined value or more is counted.

The segment correlation information is input to the person image determining section 15. The person image determination unit 15 determines the segment with the largest counted number of segments as the announcer segment.

[0005]

SUMMARY OF THE INVENTION In the above prior art,
It is possible to measure the correlation value between all the segments present in the video for a limited time. However, in the case of a long-time image, the number of segments becomes enormous, and it is very difficult to measure the correlation between all the segments and a large amount of processing time is required.

An object of the present invention is to solve the above-mentioned problems of the prior art and to provide a video search device and a search program for efficiently detecting a person in an input video.
Another object of the present invention is to provide a video search device and a search program capable of effectively detecting a person such as an announcer even from a long-time video.

[0007]

In order to achieve the above-mentioned object, the present invention provides means for inputting video data, means for dividing a segment from the video data, and motion information in the divided segment. And a means for extracting a specific color distribution in the divided segment, a means for determining a human image from the motion information and the specific color distribution, and a video search device, The point for detecting a person from the video data, or each of the above means,
The first feature is that it is made to function on a computer.

According to this feature, it becomes possible to efficiently detect a person such as an announcer in a video, and
It becomes possible to effectively detect even from a long time image.

The present invention also provides a video search device as described above.
Further, there are second, third and fourth characteristics in that a segment length measuring process, a telop detecting process and a small area color change measuring process are added respectively. According to the second feature, it is possible to speed up the person detection from the image, and according to the third and fourth features, it is possible to improve the accuracy of person detection from the image.

Further, according to the present invention, means for recording an image within a specific time, and the recorded image as image data,
A fifth aspect is that it is provided with a unit for measuring a correlation between human images detected by using one of the video retrieval devices and a unit for selecting a specific human image group using the measurement result of the correlation. There is a feature of. According to this feature, it is possible to detect a human image in a short time and with high accuracy from accumulated video data of a relatively short specific time.

[0011]

BEST MODE FOR CARRYING OUT THE INVENTION The present invention will be described in detail below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention.

In FIG. 1, it is assumed that the storage device 11 stores long-time images, for example, one-day or half-day television images (image data) recorded while the user is absent. Under the control of the control unit 10, the video input unit 12 sequentially takes out video data including a news video or the like from the storage device 11 and causes the segment detection unit 13 to input the video data.

The segment detector 13 detects the switching of the scene of the input video. Shot switching or audio scene switching, which indicates video scene switching, can be used as the segment switching. For the former, the invention of the applicant of the present invention, such as "a moving image cut screen detection method" (Japanese Patent Laid-Open No. 10-224741), can be used. Regarding the latter, an “audio information classification device”, which is an invention of the present applicant.
(Japanese Patent Laid-Open No. 10-247093) or the like can be used.

For example, it is possible to detect the time at which the voice is switched from silence to voice, or the time at which voice is switched from non-voice containing silence. Also, video cut points and audio scene switching may be combined. For example, the audio scene switching point detected after the video cut point may be set as the segment switching point. The time of the segment switching point and the video data detected by the segment detection unit 13 are input to the motion extraction unit 17, and the motion information of each segment is extracted.

As the motion information, it is possible to use the difference value between the screens of the segment top screen and the next screen. As the inter-screen difference value, for example, an inter-screen absolute difference sum of brightness values can be used. Further, a difference value between specific continuous screens in a segment or a specific screen may be used.

As a modification of the motion information extraction, an average or deviation of inter-screen difference values in a segment can be used in addition to the inter-screen difference value in a specific two screens. In this case, when a specific inter-screen difference value is small, the average or deviation of the inter-screen difference values in the segment can be obtained and used as the motion information in the segment. Further, as the inter-screen difference value, in addition to the inter-screen difference value of the entire screen, the inter-screen difference value of a specific area in the screen can be used. For example, it is also possible to divide the screen into three parts vertically and three parts horizontally and measure the difference between the screens by using only a certain area of them.

The movement information is used as the specific color distribution detection unit 18
Entered in. The specific color distribution detection unit 18 detects the specific color distribution in the segment. As a method of detecting the specific color distribution, the correlation value between the specific color distribution image and the specific image in the segment can be measured and detected. For example, it is possible to determine the correlation value by comparing the segment head image and the specific color distribution image, and determine that the specific color distribution exists when the correlation value is high. As a method of measuring the correlation value, it is possible to generate a histogram of color components of the screen and compare the histograms between the screens for measurement. As a histogram comparison method, the sum of absolute differences between screen histograms can be used. As a modification, it is also possible to use the sum of absolute differences between screens of color components.

In these correlation measurements, detection accuracy can be improved by limiting the measurement position. For example, in an announcer in a news program, only the upper half of the body appears on the screen, so the height of the face appears above the center. For this reason, when comparing the specific color distribution image and the input image, it is possible to improve the detection accuracy and the detection speed of the person by comparing at the specific position of the input image.

Furthermore, by using the correlation value at the position where the correlation with the specific color distribution image is highest among the several positions of the input image, it is possible to further improve the human image detection accuracy. For example, the correlation value can be measured with the input image at a position where the specific color distribution image is moved vertically and horizontally around the specific position. As the center of the specific position, the center value of the center of the skin-colored part of the face of the announcer segment appearing in some TV news or the position of the center of gravity can be used. As for the area to be moved, it is also possible to use the standard deviation of the center or the center of gravity of the plurality of face areas described above, or to set and move the window area based on a certain number of pixels.

As the specific color distribution image, an average of the color distributions of the faces of a plurality of persons can be used. Alternatively, a specific face image can be used as the color distribution image.

As these images, it is also possible to use those obtained by actually multiplying the input image, but it is also possible to use reduced images. Since the number of data in the reduced image is small, the processing speed can be improved. As a method of acquiring the reduced image, it is possible to use a method of periodically thinning out pixel data. Further, it is possible to easily generate a reduced image by performing frequency conversion such as DCT or wavelet and extracting only low frequency components.

In the case of a compressed moving image such as MPEG, a reduced image can be acquired at high speed by using a low frequency image. For example, in the invention “a cut point image detecting device for moving images” (Japanese Patent Laid-Open No. 9-83864) by the applicant, only a DC component is acquired from MPEG data to generate a reduced image on each screen. DC component extraction is M
Since it can be obtained by partial decoding of PEG data, it is possible to perform a high-speed image search from a compressed image.

Further, it is also possible to use only a specific screen for speeding up. For example, in MPEG, since the intra-picture coded image (I-picture) is inserted at intervals of 0.5 seconds to 1 second, it is possible to use only the intra-picture coded image. Since this screen is encoded independently of the preceding and subsequent screens, it is possible to easily generate a reduced image using only the information on this screen.

When an inter-coded image is used, the inter-coded image is a reduced image generated by using the low-frequency component of the intra-coded image, and motion information obtained by the inter-coded image. Can be generated from between the reduced images that have been reduced using. In this case, the motion vector size is converted according to the reduced image size, motion compensated prediction is performed on the reduced intra-coded image using the transformed motion vector, and the inter-coded image is displayed on the reduced image. Generate a predicted image of. Further, the DC component is extracted from the prediction error information obtained as the inter-screen encoded image data, and this D component is
By adding the C component to the preceding predicted image, a reduced image of the inter-coded image can be obtained.

The specific color distribution information obtained by the specific color distribution detecting section 18 is combined with the motion information together with the person image determining section 15
Entered in. The person image determination unit 15 determines a segment in which movement in the segment is small and a specific color distribution exists in the segment as a segment in which the person image exists. The motion within a segment can be determined to be small when the above-described difference between the two specific screens is smaller than a threshold value. Further, when there is an input image whose correlation with the specific color distribution image is equal to or more than a threshold value, it can be determined that the specific color distribution image exists in the segment.

FIG. 2 is a flow chart showing an example of the operation of the first embodiment. In this embodiment, an image is input from the storage device 11 in step S1. Step S
In 2, the segment of the input video is detected, and the process proceeds to step S3 to extract the motion information in the segment. In step S4, the motion amount in the segment is obtained from the motion information, and only when the motion amount is smaller than a certain threshold Th1, the process proceeds to step S5 and the specific color distribution detection is performed. If the amount of movement is greater than or equal to the threshold Th1, step S
The process returns to step S1 without proceeding to step 5, and the next image is input.

In step S5, if the specific color distribution is detected, the result is output to the person image determining section 15 to input the next image. If the specific color distribution is not detected, the process returns to step S1 and the next image is input.

As described above, according to this embodiment, the motion extraction unit 17 and the specific color distribution detection unit 18 are provided, and the segment in which the motion in the segment is small and the specific color distribution exists in the segment is Since it is determined that the segment includes the person image, it is possible to efficiently detect the person in the video. In addition, the step S4
If the process of (1) is provided and the process of detecting the specific color distribution is performed only when the motion in the segment is small, the person detection process can be performed in a shorter time than when the determination in step S4 is not performed. .

Next, a second embodiment of the present invention will be described with reference to the block diagram of FIG. In addition, in FIG.
The same or equivalent parts as those in FIG. 1 are designated by the same reference numerals.

The difference of this embodiment from the first embodiment is that a segment length measuring section 19 is provided at the subsequent stage of the segment detecting section 13.

FIG. 4 is a flow chart showing an example of a schematic operation of this embodiment. In this embodiment, the segment length measuring unit 19 measures the segment length, and if the segment length is larger than a predetermined second threshold Th2 (Yes in step S11), it is tentatively determined that there is a person, and Then, the process proceeds to step S3. However,
If the segment length is less than or equal to the second threshold Th2, it is determined that no person is present in the video, and step S
Return to 1 and input the next video.

As a result, in this embodiment, the person detection process can be performed in a shorter time than in the first embodiment.

Next, a third embodiment of the present invention will be described with reference to the block diagram of FIG. In addition, in FIG.
The same or equivalent parts as those in FIGS. 1 and 3 are designated by the same reference numerals.

The difference of this embodiment from the second embodiment is that a telop detection section 20 is provided after the specific color distribution detection section 18. As a telop detection method, the invention “telop area detection device in moving image” by the present applicant (Japanese Patent Application No. 2000-248794) can be used.

FIG. 6 is a flow chart showing an example of the operation of this embodiment. In this embodiment, step S1
2 is provided, and the presence or absence of a telop is used as a person detection condition. For example, in a news program, there are many images with a telop, so if there is a telop, it is determined that there is a person in the image, and if there is no telop, it is determined that there is no person. According to this embodiment, the accuracy of determining the presence / absence of a person can be improved.

Next, a fourth embodiment of the present invention will be described with reference to the block diagram of FIG. The difference of this embodiment from the third embodiment is that a small area color change measuring unit 21 is provided at the subsequent stage of the telop detecting unit 20. As the small area color change amount, a luminance component, a color difference component in the small area, or an average or deviation of the luminance and color difference components can be used. As for the small area, an area at a position related to the specific color distribution can be used.

For example, in the announcer image, the face area can be detected by detecting the specific color, and for the clothing, the area under the specific color distribution can be determined as the clothing area by using the small area color change measurement. In the determination of the small area color change, if the color change is small, that is, if the average or deviation of the color components is small, the determination condition is set.

FIG. 8 is a flow chart showing an example of the operation of the fourth embodiment. In step S13, it is determined whether or not the small area color change is smaller than a predetermined third threshold Th3. If this determination is affirmative, it is determined that a person is present in the image, and if not, a negative determination is made. If it does not exist, the process returns to step S1.

According to this embodiment, since the condition of the small area color change is added to the judgment of the person image, the accuracy of the judgment of the person image can be improved.

Next, a fifth embodiment of the present invention will be described with reference to the block diagram of FIG. The feature of this embodiment is that a specific time accumulation unit 31 is provided in the previous stage of the video input unit 12, and that a correlation measurement unit 32 and a human image group selection unit 33 are provided in the subsequent stage of the human image determination unit 15. is there.

In FIG. 9, the video input unit 12 is controlled by the control unit 10 to sequentially fetch video data such as news video from the specific time storage device 31 and input the data to the segment detection unit 13. The segment detector 13
Detects scene changes in the input video.

In the specific time storage device 31, the images within the specific time are stored. As the specific time, a specific time unit such as 30 minutes or 1 hour can be assumed. The time of the cut point detected by the segment detection unit 13 and the video data are input to the motion extraction unit 17, and the motion information of each segment is extracted. These pieces of motion information are input to the specific color distribution detection unit 18. Specific color distribution detector 18
Then, the specific color distribution in the segment is detected. The motion information and the specific color distribution information are input to the person image determination unit 15. The person image determination unit 15 determines a segment in which movement in the segment is small and a specific color distribution is present in the segment as a segment in which the person image is present.

Next, the person image segment determined by the person image determining section 15 is input to the correlation measuring section 32.
The correlation measuring unit 32 measures the correlation with each segment by using the segment head screen as the representative screen of the segment. First, the first person image segment and the second, third, ...
.., measuring the correlation with the nth person image segment. Second
For the person image segment, the correlation between the second person image segment and the third, fourth, ..., Nth person image segment is measured. As the correlation measurement, it is possible to obtain the color histogram of the images and use the sum of the absolute differences of the color histograms between the images. As a modification, it is also possible to use color histogram correlation. Further, as the representative screen of the segment, it is possible to use an image located at a position half the segment length. It is also possible to obtain the average of the color histograms of each screen in the segment and use a screen in which the sum of absolute differences between this average value and the color histogram of each screen is the minimum value.

Further, as a modified example of the correlation measurement, it is possible to update the sequential correlation measurement image. In this case, for example, the following updating method can be used.

The correlation between the first person image segment and the second person image segment is measured as described above. Next, in the case of measuring the correlation between the first person image segment and the third person image segment, the first person image is updated using the second person image. As the updating method, it is possible to use the average value of the feature amounts of the first person image and the second person image. As an example, an average color histogram of the color histogram of the first person image and the color histogram of the second person image can be used. Next, in the case of the correlation measurement of the first human image segment and the fourth human image segment, the first human image segment updated by using the second and third human image segments is used.

The above correlation value is input to the portrait image group selection unit 33. The person image group selecting unit 33 obtains an image group having a high correlation and selects the largest image group. In this case, first, an image group having a correlation value equal to or larger than a certain threshold value is obtained. The sizes of the image groups are compared, and the image group having the largest image group is output to the result output unit 16 as a person image group.

According to this embodiment, it is possible to detect a human image from accumulated video data for a relatively short specific time such as 1 hour or 2 hours in a short time with high accuracy.

Also in the above-described first to fourth embodiments, as in the fifth embodiment, the specific time storage device 31 is placed in front of the video input unit 12 to determine the person image within the specific time, and further the correlation measurement is performed. It is also possible to determine a person image group using the unit 32 and the person image group selection unit 33.

As described above, in the present invention, the human image is efficiently detected from the video recorded for a long time by performing the segment detection, the motion extraction, the specific color distribution detection, the telop detection, the brightness change measurement and the like. It was possible to detect all the segments in which the announcer appeared from the TV video recorded for 24 hours, for example. Moreover, by performing the respective processes hierarchically, it was possible to improve the processing efficiency and detect at high speed.

The processing of each of the above-described embodiments is preferably processed by software using a computer. Figure 11
FIG. 3 is a system diagram when the present invention is processed using a computer. In the figure, the same reference numerals as in FIG. 1 indicate the same or equivalent parts.

First, the video data stored in the storage device 11 is read by the video input device 12 into a video storage memory 43 such as a RAM or a cache memory that can be written and read at high speed. The main memory 42 composed of a non-volatile memory such as a ROM has a control program for issuing a command to the central processing unit (CPU) 41 and a program necessary for executing the present invention, such as a segment detection program and a motion extraction program. A specific color distribution detection program is stored. Also, if necessary,
A segment length measurement program, a telop detection program, a small area color change measurement program, etc. are stored.

Therefore, the central processing unit 41 performs segment detection from the video according to the segment detection program from the main memory 42, and then the motion information is extracted according to the motion extraction program. Next, as shown in FIG. 2, it is determined whether or not the motion amount, which is the motion information, is smaller than a certain threshold Th1, and if it is smaller, the specific color distribution is detected according to the specific color distribution detection program. Is detected. Next, the central processing unit 41 determines the segment in which the movement in the segment is small and the specific color distribution exists in the segment as the segment in which the human image exists, and outputs the result to the output device 44.

Further, the central processing unit 41 further executes the segment length measuring program, the telop detecting program, the small area color change measuring program and the like, if necessary, to obtain the segment length, the telop detecting and the small area. It is possible to add a color change and the like to the person search condition. As a result, the above-described first, second, third, fourth and fifth embodiments can be executed using a computer.

[0054]

As is apparent from the above description, according to the inventions of claims 1 to 22, the motion extracting section and the specific color distribution detecting section are provided in the subsequent stage of the segment detecting section so that the motion in the segment is small. In addition, since the segment in which the specific color distribution exists within the segment is determined as the segment in which the human image exists, the person in the video can be efficiently detected. In addition, a person such as an announcer can be effectively detected from a long-time video.

According to the second aspect of the present invention, the segment length measuring unit is further provided, and if the segment length is shorter than the predetermined length, it is determined that there is no person in the image and the next image is input. The person detection process can be performed in a shorter time.

According to the third aspect of the present invention, a telop detection unit is further provided, and if there is a telop, it is determined that there is a person in the video. If there is no telop, it is determined that there is no person. The accuracy of the presence / absence determination can be improved.

According to the invention of claim 4, a small area color change measuring section is further provided, and when the small area color change is smaller than a predetermined value, it is determined that a person is present in the image, and when the small area color change is large, it is present. Since it is determined not to do so, the accuracy of the determination of the human image can be improved.

Further, according to the sixteenth aspect of the present invention, it becomes possible to detect a human image from accumulated video data for a relatively short specific time in a short time and with high accuracy.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a first exemplary embodiment of the present invention.

FIG. 2 is a flowchart illustrating the operation of the first embodiment.

FIG. 3 is a block diagram showing a configuration of a second exemplary embodiment of the present invention.

FIG. 4 is a flowchart illustrating the operation of the second embodiment.

FIG. 5 is a block diagram showing a configuration of a third exemplary embodiment of the present invention.

FIG. 6 is a flowchart illustrating the operation of the third embodiment.

FIG. 7 is a block diagram showing a configuration of a fourth exemplary embodiment of the present invention.

FIG. 8 is a flowchart illustrating an operation of the fourth embodiment.

FIG. 9 is a block diagram showing a configuration of a fifth exemplary embodiment of the present invention.

FIG. 10 is a block diagram showing a configuration of a conventional device.

FIG. 11 is a block diagram showing a system configuration when the present invention is carried out using a computer.

FIG. 12 is an explanatory diagram of a conventional similar segment measuring method.

[Explanation of symbols]

10 ... Control unit, 11 ... Storage device, 12 ... Image input unit, 13 ... Segment detection unit, 15 ... Portrait image determination unit, 16 ... Result output unit, 17 ...・ Motion extractor, 18 ...
Specific color distribution detection unit, 19 ... Segment length measurement unit, 20 ...
..Telop detection unit, 21 ... Small area color change measurement unit, 31.
.. Specific time storage device, 32 ... Correlation measurement unit, 33 ... Person image group selection unit

Front page continuation (51) Int.Cl. ⁷ Identification code FI theme code (reference) G06T 7/20 G06T 7/20 A 5L096 H04N 5/91 H04N 5/91 N 9/79 9/79 K (72) Invention Person Hiromasa Yanagihara 2-15-1 Ohara, Kamifukuoka City, Saitama Prefecture F-term (reference) in CDI Research Institute, Inc. 5B057 AA20 CA01 CA08 CA12 CA16 CB01 CB08 CB12 CB16 CD05 CG02 DA08 DB02 DB06 DB09 DC25 5B075 ND12 NK06 NK39 5C05 AA01 AB02 AC08 CC01 DD04 5C053 FA14 GB19 GB21 HA29 5C055 CA16 EA00 FA21 5L096 AA02 AA06 BA18 FA15 GA08 GA38 HA03

Claims (24)

[Claims] 1. A means for inputting video data, a means for dividing a segment from the video data, a means for extracting motion information in the divided segment, and a specific color distribution in the divided segment. An image retrieval apparatus comprising: a hand throw for detecting; and a means for determining a person image from the motion information and the specific color distribution, and performing person detection from the image data. 2. A means for inputting video data, a means for dividing a segment from the video data, a manual pitch for measuring a segment length of the divided segment, and extraction of motion information in the divided segment. And a hand stroke for detecting a specific color distribution in the divided segment, and a hand stroke for determining a human image from the segment length, the motion information, and the specific color distribution. A video search device characterized by detecting a person from data. 3. A means for inputting video data, a means for dividing a segment from the video data, a means for measuring a segment length in the divided segment, and an extraction of motion information in the divided segment. Means, a means for detecting a specific color distribution in the divided segment, a means for detecting a telop in the divided segment, the segment length, the motion information, the specific color distribution, and the telop A device for determining a person image from a detection result, wherein a person is detected from the image data. 4. A means for inputting video data, a means for dividing a segment from the video data, a means for measuring a segment length of the divided segment, and a means for extracting motion information in the divided segment. A beat, a means for detecting a specific color distribution in the divided segment, a means for detecting a telop in the divided segment, a means for measuring a small area color change in the divided segment, Means for determining a person image from the segment length, the motion information, the specific color distribution, the telop detection result, and a small area color change, and performing person detection from the image data. Search device. 5. The video search device according to claim 1, wherein the specific color distribution is detected when a motion amount obtained from the motion information is small. . 6. The video search device according to claim 2, wherein motion information is extracted only when the segment length is large,
A video search device, wherein the specific color distribution is detected when the amount of motion obtained from the motion information is small. 7. The video search apparatus according to claim 3, wherein the motion information is extracted only when the segment length is large, and the specific color distribution of the specific color distribution is calculated when the motion amount obtained from the motion information is small. A video search device, wherein the telop is detected only when the specific color distribution is detected. 8. The video search device according to claim 4, wherein the motion information is extracted only when the segment length is large, and the specific color distribution is extracted when the motion amount obtained from the motion information is small. A video search device, wherein the telop detection is performed only when the specific color distribution is detected, and the small area color change is measured only when the telop is detected. 9. The video search device according to claim 1, wherein the means for dividing a segment from the video data uses a video cut point or an audio scene switch point as the segment switching point. A video search device characterized by being used. 10. The video search device according to claim 9, wherein a switching point between silence and voice or a switching point between non-voice and voice including silence is used as the switching point of the audio scene. Video search device. 11. The video search device according to claim 1, wherein the motion information in the segment is a difference value between frames of two specific screens in the segment, and a difference value between frames in the segment. And a motion information of only a specific partial area in the screen. 12. The video search device according to claim 11, wherein the intra-segment statistic of the inter-frame difference value in the segment is measured when the inter-frame difference value of specific two screens in the segment is small. An image retrieval device characterized by the following. 13. The video search device according to claim 1, wherein the specific color distribution is detected by using a correlation value between the specific color distribution image and the input image or a color component average value of a face area. A video search device characterized by being used. 14. The video search device according to claim 13, wherein the correlation value uses a color component histogram absolute difference sum or a color component absolute difference sum. 15. The video search device according to claim 13 or 14, wherein as the correlation value, a correlation value between a specific color distribution image and an input image at a specific position or a correlation value at a specific position of the input image and its periphery. A video search device characterized by being used. 16. The video search device according to claim 1, wherein a low frequency of frequency-converted image data is used as an input image of a unit that detects a specific color distribution in the divided segment. A video search device characterized by using image data configured by using components, or image data configured by using low-frequency components of image data obtained by partially decoding compressed image data. 17. The video search device according to claim 16, wherein reduced image data is used as the image data. 18. A means for recording a video image within a specific time period, and an interpersonal image detected by using any one of the video image retrieval devices according to any one of claims 1 to 17 as video data. An apparatus for retrieving video, comprising: a means for measuring the correlation of the above, and a means for selecting a specific person image group using the measurement result of the correlation. 19. The video search device according to claim 18, wherein the correlation between the human images is obtained using a sum of absolute differences between color component histograms. 20. The video search device according to claim 18, wherein the correlation between the person images is a correlation between segment head images,
Alternatively, a video search device characterized by using the correlation of segment center images. 21. The video search device according to claim 20, wherein the segment center image is an image at a position of half the segment length or an image closest to the average of the color histogram in the segment. Video search device. 22. The video search device according to claim 18, wherein the means for measuring the correlation between the human images uses an average of the feature amounts of the previous human images. An image retrieval device characterized by the above. 23. A computer for detecting a person from an image, means for inputting image data, means for dividing a segment from the image data, means for extracting motion information in the divided segment, And a video search program for functioning as means for detecting a specific color distribution in a segment and means for determining a human image from the motion information and the specific color distribution. 24. The video search program according to claim 23, wherein the computer further comprises a manual throw for measuring a segment length, a means for detecting a telop in the segment, and a small area color change in the segment. A video search program that functions as at least one means for measuring.