JP2000030033A - Person detecting method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To speedily detect a designated person out of pictures to be retrieved. SOLUTION: A picture including a person to be detected is designated as a key picture, the hue histogram of the face area and clothing area of the key picture is secured as retrieving data and concerning the read retrieving object picture, area dividing processing is executed to secure the hue histogram for every area, the retrieving object picture is scan-processed from two of upper and lower directions to obtain a first area with a hue histogram similar to the hue histogram of the face area and a second area with a hue histogram similar to the hue histogram of the clothing area and when the first and second areas are in a mutually proximate and vertical positional relation, this is made to be the person of a detecting object.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a person detecting method for detecting a specific person from an image to be searched.

[0002]

2. Description of the Related Art In recent years, the distribution of digital color images has increased due to the spread of digital cameras and the Internet, and the importance of the function of efficiently detecting a target image from a large amount of images has become increasingly important. Is getting higher.

[0003] Here, among the color natural images that are often seen by the user on a daily basis, snaps that are shot with a digital camera when traveling or the like are also included. Among them, the person most likely to be a subject is a person. Therefore, in a collection of images at different angles and scales,
The function of detecting an image using a specific person as a key is very useful.

As a conventional technique for detecting an image in which a person designated by a user is captured from a large number of images including a person as a subject, an area of a face or clothes characterizing the person to be searched is known. There is a technique in which a histogram is reserved in advance, a block for checking is assumed in an image, the size of the block is arbitrarily changed, and the block is compared with the histogram to search all areas of the image.

FIG. 8 shows the configuration of a conventional person detection device.
The illustrated human detection device includes a central processing unit 1, a display device 2 such as a monitor for displaying a color image, an input device 3 such as a keyboard and a mouse, and a main memory for temporarily storing programs and values being processed. Device 4 and secondary storage device 5 for storing color image data, modules for image management functions, etc.
It is composed of

[0006] The secondary storage device 5 converts the image data given by the R, G, and B values of the digital image into hue (H), luminance (V), and saturation (C) that match the human visual color difference. A color conversion unit 6 for converting to an HVC system having the following parameters: a block of an arbitrary size, a block histogram acquisition unit 16 for securing a hue (H) histogram in the block; A loop processing unit 17 for managing a loop process for checking a hue histogram in a block while changing the size and position, a face region and a clothing region for identifying a person in a hue histogram and a key image in a block at a certain point in time And a display unit 14 for displaying an image in which the degree of similarity by the matching unit 13 is higher than a threshold value predetermined by the user. It is constructed from.

Next, a person detection process for repeatedly matching the hue histograms of the face area and the clothing area of the person image specified by the key image with the search target image while moving a check block of an arbitrary size. Will be described with reference to the flowchart of FIG.

When a person image to be detected is designated as a key image (step S23), the data of each pixel is converted to an HVC system using a conversion formula from an RGB system to an HVC system (step S24).

Next, while changing the check block for any size and any place of the key image data, a block having a large number of hue skin color components is detected (step S25). When the detection is completed, the hue histogram of the face area as the block and the clothes area as the block located below the block is secured as a key image key (step S2).
6).

Thereafter, when the images to be searched are read one by one (step S27), the image data to be searched is converted into the HVC system using the conversion formula from RGB system to HVC system (step S28).

Next, similarly to step S25, a block in which many hue flesh color components are found is detected from the image to be searched (step S29). Then, similarly to step S26, the hue histogram of the face area as the block and the clothes area as the block located below the block is secured as the hue histogram information to be compared (step S30).

Thereafter, the two pieces of hue histogram information of the key image secured in step S26 are compared with the two pieces of hue histogram information of the image to be searched secured in step S30, and matching, that is, similarity is calculated ( Step S31).

If the similarity calculated in step S31 is larger than a threshold value predetermined by the user,
The image is displayed (Step S32).

[0014]

In such a conventional technique, since it is unclear at which scale and at which position a person is photographed in a certain image, the size of the person changes in the image. It is necessary to check all the positions using the blocks thus made, and to find an area where a lot of skin color components, which are considered to be a face area of a person, can be seen.

[0015] Then, there is a problem that the loop processing time is inevitably increased, and the response becomes worse when a large number of images are searched.

Accordingly, an object of the present invention is to provide a person detection method capable of quickly detecting a person specified as a key image from a search target image.

[0017]

In order to solve this problem, a person detecting method according to the present invention designates, as a key image, an image in which a person to be detected is captured, and provides a hue histogram of a face area and a clothing area of the key image. As search data,
A first hue histogram having a hue histogram similar to the hue histogram of the face region is obtained by performing a region division process on the read search target image to secure a hue histogram for each region, performing a scan process on the search target image from two upper and lower directions. A second region having a hue histogram similar to the hue histograms of the region and the clothing region is determined, and when the first region and the second region are in a close vertical relationship with each other, this is detected. It is a person.

Thus, the person specified as the key image can be quickly detected from the search target image.

[0019]

According to the first aspect of the present invention, an image of a person to be detected is designated as a key image, and a hue histogram of a face area and a clothing area of the key image is secured as search data. A first area having a hue histogram similar to the hue histogram of the face area is obtained by performing a region division process on the read search target image to secure a hue histogram for each region, performing a scan process on the search target image from upper and lower two directions. Having a hue histogram similar to the hue histograms of the
Is a person detection method in which, when the first region and the second region are in an up-and-down positional relationship close to each other, this is a person to be detected, and the person specified in the key image is Can be quickly detected from the search target image.

According to a second aspect of the present invention, in the first aspect of the present invention, there is provided a person detecting method in which a frame in which a hue histogram of the entire image sharply changes in a moving image is used as a search target image. Even if the image is a moving image, it is possible to quickly detect the person designated as the key image from a large number of search target images.

Hereinafter, an embodiment of the present invention will be described with reference to FIG.
This will be described with reference to FIG. In these drawings, the same members are denoted by the same reference numerals, and duplicate description is omitted.

FIG. 1 is a block diagram showing a configuration of a person detecting device according to an embodiment of the present invention. FIG. 2 is a flowchart showing a person detecting process in the person detecting device of FIG.
FIG. 4 is a flowchart showing a region dividing process in an image to be searched in the person detecting device in FIG. 1, FIG. 4 is a flowchart showing a person detecting process in a moving image in the person detecting device in FIG. 1, and FIG. FIG. 6 and FIG. 7 are explanatory diagrams showing a region dividing process in an image to be searched, and FIG. 6 and FIG. 7 are explanatory diagrams showing a person detecting process by the person detecting device in FIG.

As shown in FIG. 1, a person detecting device according to the present embodiment includes a central processing unit 1, a display device 2 for displaying a color image such as a monitor, an input device 3 such as a keyboard and a mouse, and a processing device. It comprises a main storage device 4 for temporarily storing programs and values therein, and a secondary storage device 5 for storing color image data, image management modules and the like.

The secondary storage device 5 is provided with a color conversion unit 6 for converting data consisting of RGB values into an HVC system having parameters matching the human visual color difference. A region division processing unit 7 for performing region division of the entire image while integrating the rectangles to be formed based on the similarity, an in-region data unit 8 for securing a histogram of hue values in the region of the image, A scanning unit 9 for examining the candidate region of the face region from the upper side of the image and the candidate region of the clothing region from the lower side of the image, and the face region candidate and the clothing region candidate detected by the scanning unit 9 are in a vertical positional relationship. Area check unit 10, a key image data unit 11 for storing data of an image of a specific person to be detected by the user, and a search for storing a large amount of image data to be searched. An elephant image data unit 12, a scene detection unit 13 which detects each transition in a frame constituting a moving image and extracts a frame having a remarkable change to detect a scene change, and a matching which calculates a similarity of a hue histogram in an area. Unit 14 and matching unit 1
And a display unit 15 for displaying on the display an image whose similarity calculated in Step 4 is larger than a threshold value predetermined by the user.

Each program stored in the secondary storage device 5 may be recorded on a recording medium such as a CD-ROM and executed by the central processing unit 1 via a reading device or the like. With such an embodiment, the present invention can be easily realized by a general-purpose computer.

Referring to FIGS. 3 and 5, a description will be given of the area division of the search target image performed in the pre-processing stage of the person detection in such a person detecting apparatus.

Here, image region division means that an image to be searched is first divided into small rectangles, and mutually adjacent rectangles are integrated based on their similarity, and the entire image is divided into its constituent regions. Is what you do.

In the area division of an image, first, image data to be divided is secured (step S12), and this image data is converted from the RGB system to the HVC system (step S13). Then, the entire image is divided into rectangles of a predetermined size (step S14), and for all the divided rectangles, histogram information of the hue (H of the HVC system) in the block is obtained (step S15). ).

Next, the similarity of the hue histogram is calculated for rectangles having a positional relationship adjacent to each other, such as up, down, left, and right, and the obtained similarity is smaller than a threshold determined by the user in advance. If it is larger, those small rectangles are regarded as the same area and integrated (step S16).

Finally, a so-called labeling process for giving a unique label to the region integrated in this way is performed (step S17).

Such a series of processing will be described more specifically with reference to FIG. Images to be searched (Fig. 5
Given (a)), this image is divided into small rectangles, and a histogram within the rectangle is secured (FIG. 5).
(B)).

Then, the hue histogram information, which can be said to be the feature of each small rectangle, is integrated by mutually similar ones, and a region constituting the entire image is extracted (FIG. 5C).

As described above, if an image is divided into small rectangles and these small rectangles are integrated in accordance with the similarity, a certain degree of accuracy can be obtained for the image to be searched in a rough region characterizing the image. Can be divided by

Next, the details of the person detecting process by the person detecting apparatus according to the present embodiment will be described with reference to FIG.

In FIG. 2, when an image including a person as a key for detection is designated by a user as a key image (step S1), if the image having data in the RGB system matches the visual color difference of a human, H having parameters of hue (H), luminance (V), and saturation (C)
It is converted to a VC system (step S2).

The image converted to the HVC system is divided into small rectangles, and a histogram for each small rectangle is calculated based on the hue data obtained in step S2. Then, if the similarity between mutually adjacent rectangles is larger than a threshold value predetermined by the user, the two are integrated, and by performing such processing on all the rectangles, the region division processing is performed ( Step S3).

After the area division processing is performed on the key image specified by the user, the hue histogram of the area having a large number of skin color components in the hue data information and the hue histogram of the clothes located below the area are displayed in the key image. Secured as search data (step S4).

When the search data of the image with the key is secured in this way, the image data to be searched is read (step S5). Then, as in step S2 described above, the image having data in the RGB system has parameters of hue (H), luminance (V), and saturation (C), which are said to match human visual color differences. It is converted to an HVC system (step S6).

Next, an area dividing process is performed on the search target image in the same manner as in step S3 (step S3).
7), a hue histogram for each area is secured (step S8).

Then, scan processing is performed on the search target image, and the histogram for each area is sequentially checked from the upper and lower two directions (step S9).

In step S9, if an area scanned from above is found to have a large amount of skin color components, the result of the processing scanned from below is checked, and an area similar to the hue histogram of the clothes of the key image exists. A matching process is performed to check whether the region where the region exists is in a vertical positional relationship with the face region candidate (step S10).

Finally, based on the similarity checked in step S10, an image in which a person in the key image is detected is detected from a large number of images to be searched and displayed on the display device (step S11). .

Such a series of processing is described in FIG. 6 and FIG.
This will be specifically described with reference to FIG. Image to be searched (Fig. 6
Given (a)), this image is divided into small rectangles. Then, they are integrated based on the similarity of the hue histograms of the respective rectangles, and are divided into rough regions characterizing the image (FIG. 6B). Here, for example, A, B, C, D,
It is divided into six areas EF.

When the area is divided, many small areas appear. However, since only what is necessary here is an area having a certain size that is considered to be judged when viewed by a human, a fine area is excluded from processing.

After the area division is performed as described above, FIG.
As shown in (a), while moving the scan line from top to bottom, an area having a hue histogram in which many flesh color components are seen is searched for.

Here, in the scanning process, first, the hue histogram of the A, B, and C regions that are in contact with the upper side is checked. Then, the scan line is moved to the lowest position of these checked areas, and the following areas are sequentially checked.

When the scan is performed in this way and an area that is considered as a face area candidate appears, the movement of the scan line from above is stopped when the scan line comes to the end of the area. In the case shown in FIG. 7A, since many skin color components are found in the area B, the scan line is stopped at a position where the scan line has moved to the bottom of the area B.

Next, as shown in FIG. 7 (b), the hue histogram of the clothing area included in the person image which is the key image designated by the user in advance while moving the scan line upward from below. Check for areas that are similar to. Here, D, which is an area in contact with the lower part of the image
-Acquire the information of the hue histogram for E and compare it with the hue histogram of the clothing area of the key image.

Then, the scan line is moved while calculating the similarity between the key image data and the hue histogram of the area, and stopped at the position where the scan line has been moved to the top of the similar area. Here, since the region E is regarded as a clothing region, the scan line is stopped at a position where the scan line has moved to the top of the region E.

Here, the distance ΔD between the final position of the scan line from the upper direction where the area B is checked and the scan line from the lower direction where the area E is checked (FIG. 7C).
Is calculated. If the interval ΔD is within an allowable range predetermined by the user, since the face area and the clothing area of the person have a vertical positional relationship close to each other, these candidate areas are face areas (B). And the clothing area (E), it is specified that the user is a person image given as a key image.

In the above description, a still image is a search target, but the present invention can be applied to a moving image. this is,
This is because a frame constituting a moving image can be processed as a still image.

Here, the person detection process in a moving image will be described with reference to the flowchart of FIG. In a moving image, two major processes, scene detection and person detection, are performed in order to detect a frame in which a person designated as a key image by a user is shown.

That is, when moving image data serving as a search target image is read (step S18), an entire hue histogram is obtained for all frames of the image (step S19).

Next, as the inter-frame comparison processing, a point where the hue histogram changes abruptly in frames preceding and succeeding in time series is checked (step S20), and the frame at the checked point is detected (step S21).

The frame detected in step 21 corresponds to a change of a scene in a moving image. Therefore, a person detection process is performed on the frame detected in step S21 (step S22).

Thus, a scene including a person designated by the user as a key image can be detected from a moving image.

As described above, according to the present embodiment, regardless of whether the detection target image is a still image or a moving image, the person designated as the key image is quickly detected from a large number of search target images. It becomes possible to do.

[0058]

As described above, according to the present invention, it is possible to obtain an effective effect that the person designated as the key image can be quickly detected from the search target image.

Further, if a frame in which the hue histogram of the entire image changes rapidly in a moving image is set as a search target image, even if the detection target image is a moving image, a person designated as a key image can be searched for a large amount of An effective effect that the detection can be quickly performed from the inside is obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a person detection device according to an embodiment of the present invention;

FIG. 2 is a flowchart showing a person detection process in the person detection device of FIG. 1;

FIG. 3 is a flowchart showing an area dividing process in an image to be searched in the person detecting apparatus in FIG. 1;

FIG. 4 is a flowchart showing a person detection process in a moving image in the person detection device of FIG. 1;

FIG. 5 is an explanatory diagram showing an area dividing process in an image to be searched in the person detection device in FIG.

FIG. 6 is an explanatory diagram illustrating a person detection process performed by the person detection device of FIG. 1;

FIG. 7 is an explanatory diagram showing a detection process of a person by the person detection device of FIG. 1 subsequent to FIG. 6;

FIG. 8 is a block diagram showing a configuration of a conventional person detection device.

FIG. 9 is a flowchart showing a process of detecting a person by the person detecting apparatus of FIG. 8;

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 Central processing unit 2 Display device 3 Input device 4 Main storage device 5 Secondary storage device 6 Color conversion unit 7 Area division processing unit 8 In-area data unit 9 Scanning unit 10 Area boundary check unit 11 Key image data unit 12 Search target image Data section 13 Scene detection section 14 Matching section 15 Display section

Continued on the front page F-term (reference) 5B050 AA09 BA12 DA04 EA08 EA09 FA02 FA05 5B057 AA01 BA02 CA01 CA08 CB01 CB08 CB11 CB12 CC01 CE16 DA08 DB02 DB06 DB09 DC14 DC19 DC25 DC32 5L096 AA02 AA06 BA02 CA02 DA45

Claims (2)

[Claims] An image including a person to be detected is designated as a key image, hue histograms of a face region and a clothing region of the key image are secured as search data, and a region division process is performed on the read search target image. The hue histogram of each area is secured, and the search target image is scanned from the upper and lower two directions, and the hue histogram of the first area and the hue histogram of the clothes area similar to the hue histogram of the face area are obtained. A second area having a hue histogram is obtained, and when the first area and the second area are in a vertical positional relationship close to each other, this is set as a detection target person. Person detection method. 2. The method according to claim 1, wherein a frame in which the hue histogram of the whole image changes rapidly in the moving image is used as the search target image.