KR101329138B1 - Imaging system, apparatus and method of discriminative color features extraction thereof - Google Patents

Abstract

An image forming system, a color feature extraction apparatus having strong discriminating power, and a method thereof are provided. According to the present invention, a photographing apparatus for photographing an image of an object, receiving an image from the photographing apparatus, extracting a color feature having strong discriminating power from the image, generating a final color model of the object, and extracting a color blob of the object based on the final color model A color feature extracting device for generating a variable for adjusting a state of the photographing apparatus according to the analysis, and receiving a variable for adjusting a state of the photographing unit from a color feature extracting device and adjusting a state of the photographing unit Apparatus, a storage device for storing an image of a photographed object, and a display device for displaying an image of the photographed object.

Image, discernment, empty, blob, color model

Description

Image forming system, apparatus for extracting color features having strong discrimination ability and method thereof {IMAGING SYSTEM, APPARATUS AND METHOD OF DISCRIMINATIVE COLOR FEATURES EXTRACTION THEREOF}

The present invention relates to pattern identification, feature extraction, statistical learning, and object detection techniques (generally, human body detection techniques), and more particularly to extraction of features having a strong discrimination ability with respect to a target color or other objects of the same or different kind. An image forming system for forming a color distribution pattern having a strong strong discriminating power of the present invention, a color feature extraction apparatus having a strong discriminating power used therein, and a method thereof.

Representative color features of object appearance have been widely used for object detection and tracking of video frames. Color histograms are typically used to represent representative color features. In order to know the spatial distribution of the color distribution, it is suitable to use a histogram of the divided independent regions of the target region. In addition to this method, a color model can be corrected by another method, for example, motion information, shape information, geometric constraints, or the like, to obtain better detection results.

Several attempts have already been made to solve the problem of the associated exercise object using the color model. These attempts can be accomplished using probable tracking using color appearance, forming color histogram models and detecting probability plots, comparing multiple similar search boxes based on color models, extracting skin color based on statistical models, etc. There is this.

Representative color features of object appearance are widely used for object detection in video frames. When displaying color features, a color histogram is generally used to determine the position and size of the exercise object and to calculate the color histogram of the exercise object. The similarity event of the histogram model is calculated in the next video frame. Then, pixel regions having high similarity to the histogram model of the blob analysis technique can be collected and indicated by position blobs of the moving object of high probability.

Note how many methods efficiently display the color characteristics of the exercise object. A practically important problem is how to distinguish between the exercise object and the non-movement object, ie other objects and backgrounds of the same or different kind, but not to faithfully and perfect the original color distribution of the object. For exercise subjects this consists of a number of parts. For example, if the person to be exercised is a person, this person includes a face / head, an upper body of the complexion and a lower body of the complexion. The color appearance of a person may perhaps be similar to other image areas, especially under complex background environments. In such a situation, it is necessary to select a color gamut having strong discriminating power from the human body as a color feature, but does not select a representative color feature of the entire human body.

Strongly distinguishable color features represent large color discrimination that exists between the motion object itself and other image areas, including background images. In order to extract color features having a strong discriminating power, the color vectors of the color histograms of the plurality of divided objects and the color vectors of the color histograms of the other image areas are compared to each other. Lower the importance of color vectors and increase the importance of other color vectors that differentiate.

Accordingly, the present invention provides an image forming system for extracting color features having a strong discriminating power for image forming and a different environment of a scene, and providing a focus area for adjusting the state of the image forming unit, a color feature extracting device having a strong discriminating power thereof, and It aims to provide the method.

In order to solve the above problems, the present invention, the photographing apparatus for photographing the image of the object, and receives the image from the photographing apparatus and extracts a color feature having a strong discrimination power from the image to generate a final color model of the object And a color feature extraction device for extracting a color blob of the object based on the final color model to perform analysis and generating a parameter for adjusting a state of the photographing device according to the analysis. And a control device for receiving a variable for adjusting the state and adjusting the state of the photographing unit, a storage device for storing the image of the photographed object, and a display device for displaying the image of the photographed object. Provide an image forming system.

In order to solve the above problems, the present invention also divides the target area and the non-target area of the input image into a plurality of rectangles to form a color histogram for each color channel and extracts the main color of the target rectangle from the color histogram. A color extraction unit, a minimum distance between the color histogram bin of the target rectangle and the color histogram bin of the non-target rectangle with respect to the main color, the bin weight of the color histogram of the target rectangle, the color channel weight of the target rectangle, and A weight calculation unit configured to calculate weights of the target rectangles to extract color features having a strong discriminating power of the targets, and generate a final color model of the targets according to a result of re-weighting the color histogram bin of the target rectangles; Having a final color model generation unit It provides a color feature extraction apparatus as set.

In order to solve the above problems, the present invention also comprises the steps of dividing the target area and the non-target area of the input image into a plurality of rectangles to form a color histogram for each color channel and extracting the main color of the target rectangle from the color histogram. And calculating a minimum distance between the color histogram bin of the target rectangle and the color histogram bin of the non-target rectangle with respect to the main color, and calculating the bin weight of the color histogram of the target rectangle based on the minimum distance. Calculating a color channel weight of the target rectangle and a weight of the target rectangle based on the bin weights; and weighting the color histogram bin of the target rectangle again based on the color channel weight and the weight of the target rectangle. Strong discriminating power of the subject Having provides a color feature extracting method comprising the step of generating the final color model of the object based on the result is added a step of extracting a color feature and, again, the weight to the color histogram bin of the target rectangle.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. Here, although the present invention has been described with the object of exercise as an object, the present invention is not limited thereto and may be applied to other types of exercise objects.

1 is a block diagram of an image forming system according to an embodiment of the present invention.

The image forming system 900 includes a photographing apparatus 901, a color feature extraction apparatus 902, a control apparatus 903, a storage device 904, a display apparatus 905, and a notation for extracting color features having strong discriminating power. Device 906. The image forming system may be any one of a PTZ (PAN, TILT, ZOOM) camera, a still surveillance camera, a DC (digital camera), a DV (digital video), and a PVR (personal video recorder).

The imaging device 901 may be a hardware device, for example, a CCD or COMS device, used for sensing and generating a natural image, and the image processing chip of the imaging device 901 may exhibit good image quality. In order to track an exercise target, two kinds of methods may be used to provide the position and size of the target exercise region.

The first method is an automatic method, which uses an embedded algorithm to extract the size and location of the object of interest. The second method is a manual method, in which a user or an operator labels an area of interest on an displayed image (for example, a touch screen). The automated method can automatically detect the object using an embedded algorithm. The marking device 906 provides a marking function to the user or operator so that the user or operator can manually mark the region of interest on the image with a pen or finger.

The color feature extraction apparatus 902 having a strong discrimination power receives image data from the photographing apparatus 901, and also receives size and position information of the region of interest marked in a rough notation form from a user. The color feature extraction apparatus 902 having a strong discrimination power extracts a color feature having a strong discrimination power in a region other than the object to be tracked, that is, the remaining region including the background, generates a final color model of the object, and generates a final color. Based on the model, the color blobs of the object are extracted, and the color blobs are analyzed, and the color blobs are analyzed to generate variables for adjusting the state of the image forming system. Note that the marking device 906 may be selected when using the first method of providing the target location and size. When selecting multiple tracking targets, and when there are multiple tracking targets, the user may modify the automatic selection tracking target of the image forming system 900 of the present invention.

The control device 903 adjusts the state of the image forming system 900 and selects a still surveillance camera, DC, DV or the like by selecting a pan, tilt, zoom and focusing area of the PTZ camera. Automatic condensing operation control or zoom operation control of the PVR is performed. The control device 903 receives a variable for adjusting the state of the image forming system 900 from the color feature extraction device 902 having strong discriminating power. The color feature extraction apparatus 902 having a strong discrimination power may provide target position and size information of a new time point or new frame data. The control device 903 adjusts the state of the image forming system 900 based on the variable so that the object is positioned in the middle of the image through the rotation or tilting operation, and selects the region of interest through the manipulation of the light converging area. By focusing on the region of interest through zooming and automatic focusing operation, it is possible to obtain details of an exercise target with high image quality. During the operation of selecting the condensing area, the control device 903 can condense the area by controlling the image forming system 900 to select a new area to be the object of condensation.

In addition, when the image forming system 900 selects the condensing region, the control device 903 selects a new object in which the object is dynamically located in addition to the image forming system 900 selecting the image center region as the image forming condensing region by default. The image area can be selected as the image forming focusing area, and based on the image data information of the focusing area, a better image is obtained by dynamically adjusting the zoom multiple, auto focus, focus, rotation, or tilt parameters of the image forming system 900. The formation effect can be obtained.

For electronic products on the market, such as DC, DV or PVR, the user can position the object of interest in the middle of the image by manually adjusting its state. At this time, the device itself automatically performs other operations, such as zoom multiple correction and auto focus.

The storage device 904 can store an image or a video in the storage device, and the display device 905 can display an image or a video of the field to a user.

The present invention may be realized in software, which may be used to connect a system embedded in the image forming system 900 and the control device 903 to adjust the parameters of the image forming system state. By receiving video as an input to the embedded image forming system 900 and sending a command to the control device 903 of the image forming system, the state of the image forming system 900, the mirror condensing area, and the like can be adjusted.

The present invention is not intended to extract the target color appearance as a representative color histogram, but to extract a color model having strong discrimination power. There is a difference between a representative color model and a color model with strong discrimination. Sometimes when there is a large difference between the color appearance and non-object and background color models of the exercise object, the color model having a strong discrimination power and the representative color model may be similar. Sometimes when the color model having strong discriminating power and the representative color model are dissimilar, especially when there is some difference in the color appearance of the exercise object and the color model of the non-object and the background, the present invention effectively prevents interference colors having similarities to the exercise object. The target object can be positioned using a color model having a strong discriminating power among the complex background of the region.

According to the present invention, a good image forming condition can be obtained by adjusting an image forming system by distinguishing a plurality of exercise objects and a tracking object having a partly similar color from each other, and subsequently determining the position and size of the tracking object. Color models with strong discrimination use the importance weighting strategy principle. Color models with strong discernment increase the importance, or weight, of color components that differ significantly from non-objects or backgrounds, and reduce the importance, or weight, of color components similar to non-objects or backgrounds.

Using a still camera or tracking an object with a PTZ camera of a target exercise target that is used or suspected of interest may be used for event analysis and image quality improvement. In detection using a color model with strong discrimination, the detection scans the input image and then converts the input image into a blob image based on the color model with strong discrimination. By analyzing the blob images it can be used to detect human candidates from human body detectors to eliminate error warnings.

Among the PTZ cameras, the present invention is realized by software in an embedded system, which receives an input video image and analyzes the input image to form a color model having strong discriminating power of the input image, and then positions the person. By tracking the person by adjusting the state of the camera (rotation, tilting, zooming, condensing area selection), the person can be placed at the center of the image, and can exhibit good fractionation and image quality. When the camera is rotated and tilted, tracking is performed on the person who is exercising so that the person does not leave the field of view of the camera. When optically zooming the camera, high quality images can be obtained for the local or entire body of the person exercising by adjusting the optical focus. The video recording system can store field data in a storage device, and the video recording system can record human actions or interact with the device at high image quality as it enters the area of the user device.

Note that the present invention can be used in portable image forming systems, such as DC, DV or mobile cameras. The initial position and size of a person is indicated manually by the user or automatically detected by a detector. By directly positioning and focusing the exerciser, the imager can automatically focus on the exerciser to achieve high image quality. The terminal user manually adjusts the mirror direction of the image forming system of the person so that the person who is exercising is located in the middle of the image.

In still cameras, the present invention can implement a set of trajectories of the exerciser associatively tracking the exerciser. The trajectory includes the position and size of the person exercising. During the tracking, the system analyzes what happened or whether a series of warnings or notifications were sent out. The user-defined event may be a dangerous situation or a general notification situation of the user, for example, the position or size of the exerciser may be an intruder who has entered a certain pre-installed prohibited area.

The term "image forming system" used in the present invention is a device having at least an image device and a control device, and may be a camcorder, a camera, or other portable device having an image forming function.

The present invention can be used in robots that position a person using a robotic image forming system. Basic functions of the robot may include avoiding obstacles, interacting with people, tracking people, and the like. The present invention can track or interact with a person by specifying whether a person exists in a nearby area or a location of the person.

The present invention can also be used for the extraction and detection of color features with strong discrimination of other types of athletic objects, which has been described above by way of example. Application to candidate goals is not limited to human modeling and detection. For example, an image is formed on the exercise target using DC, and the terminal user may indicate the target position, size, or region by manually marking a region of interest by drawing a line on the touch screen with a hand or a pencil. The present invention can form a model with a strong discriminant of an object and indicate the position and size of the video frame below the object. Therefore, the DC can obtain high image quality of the exercise object by adjusting the image forming parameters.

2 is a block diagram of an apparatus for extracting color features having strong discrimination power according to the present invention.

Referring to FIG. 2, the color feature extraction apparatus 902 having strong discrimination power according to the present invention includes an area divider 801, a histogram calculator 802, a main color extractor 803, and an empty minimum distance calculator. 804, the bin weight calculator 805, the color channel weight calculator 806, the rectangular weight calculator 807, the bin weight recalculator 808, and the final color model generation unit 809, and separately include a target. The location unit may further include.

Looking back, the color feature extraction apparatus 902 according to an embodiment of the present invention divides the target area and the non-target area of the input image into a plurality of rectangles to form a color histogram for each color channel, and the target from the color histogram. A main color extraction unit 880 for extracting a main color of a rectangle, a minimum distance between the color histogram bin of the target rectangle and the color histogram bin of the non-target rectangle with respect to the main color, and a color histogram of the target rectangle A weight calculation unit 881 for calculating a bin weight, a color channel weight of the target rectangle, and a weight of the target rectangle to extract color features having a strong discriminating power of the object, and weighting the weight back to the color histogram bin of the target rectangle; Generate a final color model that generates the final color model of the object according to the result Unit 809 is provided.

The main color extraction unit 880 extracts a main color for extracting the main color of the target rectangle based on the color histogram formed by the area divider 801, the histogram calculator 802, and the histogram calculator 802. The part 803 is provided.

The area divider 801 divides the target area and the non-target area in the image into a plurality of rectangles. The histogram calculator 802 forms a color histogram of a target rectangle and a non-target rectangle using a color channel, and the color histogram is divided into a plurality of separated bins. The main color extraction unit 803 extracts the main color of the target rectangle based on the color histogram.

The weight calculator 881 includes a bin minimum distance calculator 804, a bin weight calculator 805, a color channel weight calculator 806, a rectangular weight calculator 807, and a bin weight recalculator 808. do.

The bin minimum distance calculator 804 calculates a minimum distance between the color histogram bin of the target rectangle and the color histogram bin of the non-target rectangle with respect to the main color extracted by the main color extraction unit 803. This is to reduce the weight of the bin whose color discrimination with the non-target histogram bin among the target color histogram bins. The empty minimum distance calculator 804 calculates the minimum distance according to equations (1) and (2).

는 대상 직사각형(r)의 색 채널(c)의 빈(i)의 색 히스토그램이고,

는 색 채널(c)의 빈의 비대상 색 히스토그램이며,

는 빈의 표지이고,

는 대상 직사각형(r)의 색 채널(c)의 빈(i)와 동일한 색 채널을 가지는 비대상 직사각형 색 히스토그램 빈과의 최소 거리를 나타낸다.Where J denotes the bin of the color histogram of the non-target rectangle of the color channel (c), and the bin (J) is the minimum with the bin (i) of the color channel (c) of the main color histogram of the target rectangle (r). Distance, bi is the sign of bin (i).

Is the color histogram of the bin (i) of the color channel (c) of the target rectangle (r),

Is the non-target color histogram of the bins in the color channel (c),

Is the cover of the bin,

Denotes the minimum distance from the non-target rectangular color histogram bin having the same color channel as the bin (i) of the color channel (c) of the target rectangle (r).

The bin weight calculator 805 calculates a weight of the color histogram bin of the target rectangle based on the calculated minimum distance of the bin minimum distance calculator 804. The bin weight calculator 805 calculates the weight of the color histogram bin of the target rectangle according to Equation (3).

here,

k는 상수here,

k is a constant

The color channel weight calculator 806 calculates the weight of the color channel of the target rectangle based on the weight of the color histogram bin of the target rectangle calculated by the bin weight recalculator 805. The color channel weight calculator 806 calculates the weight of the color channel of the target rectangle according to equations (4) and (5). This is because the distance from the non-target histogram bin among the bins of the target histogram increases the weight of the bin on the far side and the weight of the close bin to reduce the histogram of the object.

In Equation 4, the weight of each bin of the color channel c is accumulated. Therefore, the weight of each color channel is calculated according to Equation 5. In Equations 4 and 5, M denotes the number of bins of a human rectangle, and K denotes the number of color channels.

The rectangular weight calculator 807 calculates the weight of the target rectangle based on the weights of the color histogram bins of the target rectangle calculated by the color channel weight calculator 806. The rectangular weight calculator 807 calculates the weight of the target rectangle according to equations (6) and (7).

Calculate the bin weights of the human rectangle (r) in equation (6), calculate the weights of the rectangles in equation (7), and equation (7) proceed the importance weights for these rectangles of the color bins with high discernibility, where L Represents the number of people rectangles.

The bin weight recalculation unit 808 calculates the color histogram bin of the target rectangle based on the weight of the target rectangle calculated by the rectangular weight calculator 807 and the weight of the color channel calculated by the color channel weight calculator 806. New weights are applied to extract color features having a strong discriminating power of the object. The bin weight recalculation unit 808 newly weights the color histogram bin of the target rectangle according to Equation (8).

The final color model generation unit 809 generates the final color model of the object based on the newly weighted result of the color histogram bin of the object rectangle. The final color model generation unit 809 generates the final color model of the object according to the following equation:

Where R denotes the number of target rectangles, and M denotes the number of bins of the target rectangle histogram.

In addition, the apparatus of color feature with strong discriminating power extracted in accordance with the present invention also includes a color blob extraction unit 810 and a target position unit 811.

The color blob extraction unit 810 extracts the color blobs of the object from the newly input image based on the final color model of the object. The target position unit 811 performs color blob analysis on the color blob extracted by the color blob extraction unit 810, calculates the shape center and size of the object in the input image, and performs positioning and tracking on the object. do.

3 is a flowchart of a method for extracting color features having strong discrimination power in accordance with the present invention.

In step 101 the system receives a video input, the location and size information of the person, the location and size information of the person being given via exercise extraction, manual display or human body detector.

There are many methods of motion extraction, for example background subtraction can be used since the initial state of the image forming system is stationary. Manual notation can be used to give a person's position and size information directly. The human body detector detects an input image used for the position and size of the candidate person. Although people in the image have different sizes, the ratio of width and height has a fixed range and can be used to securely determine other image areas other than the exercise object. That is, the non-target area includes other objects and backgrounds of the same or different kind.

The next step is how to extract the color gamut with strong discrimination based on the known rectangular and non-human rectangular image areas. Quantization is performed on the color image in step 102, where the color pixels having three color channels are quantified into N scattered bins, for example N = 64.

In step 103, the rectangles of the human head and the human body are calculated. 2 shows the calculations for the rectangles of the human head and human body. For example, when determining the position and size of a person's head, several rectangles of the same size are arranged downward and proceed to the lower boundary of the image. As is already known, a person's face and a person's head have a distinct color, and the head's rectangle is divided into several small rectangles. According to the same principle, image areas other than the area occupied by a person can be safely measured based on the ratio of the height (head) width of the person in the statistics. It is not necessary to accurately measure the background rectangular area, i.e., it is possible to roughly measure the background rectangular area, so that an image pixel without a moving object can be classified as a background rectangle.

In step 104, a rectangular histogram is calculated. FIG. 5 shows the calculation of the color histogram of each rectangle, which comprises a head rectangle, a human rectangle and a background rectangle (or other rectangles other than the given human rectangle, ie non-target area). These rectangular histograms are each calculated, the histogram is divided into a number of bins, and the number of pixels in each bin is increased.

In step 105, after obtaining the color histogram of each rectangle, the main color is preserved and the secondary color is deleted. Since noise can affect the secondary color with a small amount of pixels, and such pixels have the secondary color, the secondary color is unstable or unreliable.

In step 106, the minimum distance of the bins is calculated.

As shown in Equations 1 and 2, the distance between the rectangular color histogram bin of the object and the non-object is calculated. The example below shows the background area as a non-target area.

Step 107 calculates the weight of the bin.

As shown in Equation 3, the discriminating power (weighting value) of one bin of the rectangle r and the channel c is defined.

In step 108 the weights of the color channels are calculated.

In each rectangle, the pixels consist of three color channels, each of which is a red, green, and blue component. Each type of component is quantified into a number of histogram bins so that each component determines the distance between the histogram bins with respect to them and has a different discrimination power.

In step 109 a rectangular weight is calculated.

The importance of a person's rectangle is determined by the discernment of the color bins it contains. When a rectangle contains a high color gamut, the rectangle has a greater weight than the other rectangles.

In step 110, the histogram bin is newly weighted.

, 색 채널 가중치

및, 직사각형 가중치

로부터 결정된다. 따라서, 히스토그램의 최종 중요성은 수학식 9에 따 라 다시 가중치가 정해진다.Each histogram bin has a different discrimination between the target exercise object and the complex background or other exercise object. The significance of the histogram bin is the initial weight of the histogram bin of the histogram.

, Color channel weights

And, rectangular weights

Is determined from. Therefore, the final importance of the histogram is weighted again according to equation (9).

In step 111 the final color model is generated.

는 사람 머리 직사각형의 개수,

는 사람 신체의 직사각형의 개수,

는 사람 머리의 직사각형 히스토그램의 빈 개수,

는 사람 신체 직사각형 히스 토그램 빈의 개수를 각각 표시한다.The histogram bins of each human rectangle are weighted again, and then a final color model is generated according to equations 10 and 11, which represent the final color model (color histogram distribution) of the human head and human body, respectively. do. In Equations 10 and 11, the weights of the histograms of the rectangles of the human head and the body are accumulated, respectively. R is the number of human rectangles, and M is the number of bins of human rectangles. In detail,

Is the number of people head rectangles,

Is the number of rectangles in the human body,

Is the empty number of rectangular histograms of a person's head,

Denotes the number of human body rectangular histogram bins respectively.

To illustrate the main idea later in this section, FIG. 6 shows an embodiment.

Fig. 6A is a color histogram of one rectangular color channel of the exercise object. FIG. 6B is a color histogram of one color channel among other areas of an image other than an exercise object. There are two bins of the two color histograms, indicated by BIN-1 and BIN-2. As BIN-1 of the exercise target is farther away from the two bins in the background than BIN-2 of the exercise target in the distance transformation (Equations 1 and 2), as shown in FIG. The weight of BIN-1 increases while the weight of exercise target BIN-2 decreases. 6 shows the importance weighting function T (x, y), which not only increases the weight having a large bin distance but also decreases the weight having a small bin distance so that the bin distances are represented by Equations 1 and 2 Is calculated according to.

Note that here we show how to display the three color channels for a color image, one of which uses a one-dimensional color histogram to form the final color model independently. However, there are many changes in practice, using two-dimensional color histograms or three-dimensional color histograms. For example, there are three kinds of two-dimensional color histograms, a red-green two-dimensional color histogram, a red-blue two-dimensional histogram, and a blue-green two-dimensional histogram. However, the method of the present invention also extracts color features with strong discrimination by using such histograms in a similar manner.

Based on the final color model extracted and generated from the color features having strong discriminating power, a color blob belonging to the motion object is extracted from the new input image. For each pixel in the input image, the similarity of the target motion object is formed by a weighting matrix, and the size of the weighting matrix is the same as that of the input image. The components of the weighting matrix come from the accumulated weights of the lowest color histogram model and the final color histogram model can obtain quantization using color pixel values.

In Figure 8, there are two people. One person is selected for tracking according to some kind of method, for example, the person on the right is selected for tracking. 8A is a new input image, FIG. 8B is a result of blob detection of the user body model, and FIG. 8C is a result of blob detection of the user head model. White is widely distributed in the background and others, so the white of the right person has a low weight and accounts for most of the initial weight of the color histogram.

Referring to FIG. 9A, there are two head candidates from a human head detector having a shape model. Thus, the probability distribution of the head is a Gaussian distribution around a given head position. 9 (b) shows the result of color blob detection using the hair color model, and FIG. 9 (c) shows the result of color link detection using the body color model. In the detection of color blobs in the head, the probability distribution of a person is a Gaussian distribution around a given head position.

Assuming a person's up and down posture, the probability distribution extends downward, indicating the presence of the human body. In the detection of human body blobs, the probability distribution of humans is a Gaussian distribution around a given body position. Since the position of the human head is in the parietal region of the body (which constrains the geometric relationship), the probability distribution extends upwards to indicate the presence of the human head. Therefore, referring to FIG. 9D, the shape center of the probability distribution may be calculated as the final position of the tracking target.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, Modification is possible. Accordingly, the spirit of the present invention should be understood only in accordance with the following claims, and all equivalents or equivalent variations thereof are included in the scope of the present invention.

1 is a block diagram showing an image forming system according to the present invention.

2 is a block diagram showing an apparatus for extracting color features having strong discriminating power in accordance with the present invention;

3 is a flow chart illustrating a method of extracting color features having strong discriminating power in accordance with the present invention;

4 illustrates a method of calculating a human rectangle;

5 shows the calculation for each rectangular histogram,

6 is a view showing the calculation of the bin weight and the color extraction effect having a strong discrimination power using the present invention,

7 is a diagram showing a function to which importance weights are applied;

8 shows detection of color blobs using color features with strong discriminating power,

9 shows a blob analysis used to determine the position and size of a person,

Claims (25)

delete delete delete delete delete A main color extraction unit for dividing the target area and the non-target area of the input image into a plurality of rectangles to form a color histogram for each color channel and extracting the main color of the target rectangle from the color histogram; The minimum distance between the color histogram bin of the target rectangle and the color histogram bin of the non-target rectangle with respect to the main color, the bin weight of the color histogram of the target rectangle, the color channel weight of the target rectangle, and A weight calculation unit for calculating a weight to extract color features having a strong discriminating power of the object; And A final color model generation unit for generating a final color model of the object according to a result of weighting the color histogram bin of the object rectangle again; Color feature extraction apparatus characterized in that it comprises a. The method of claim 6, wherein the main color extraction unit An area divider dividing a target area and a non-target area of the input image into a plurality of rectangles; A histogram calculator for forming a color histogram of the object rectangle and the non-object rectangle using a color channel and dividing the color histogram into a plurality of separated bins; And A main color extracting unit extracting a main color of the object rectangle based on the color histogram formed by the histogram calculating unit; Color feature extraction apparatus characterized in that it comprises a. The method of claim 6, wherein the weight calculation unit, A bin minimum distance calculator configured to calculate a minimum distance between the color histogram bin of the target rectangle and the color histogram bin of the non-target rectangle with respect to the main color; A bin weight calculator configured to calculate a weight of the color histogram bin of the target rectangle according to the minimum distance calculated by the bin minimum distance calculator; A color channel weight calculator configured to calculate weights of the color channels of the target rectangle based on the bin weights calculated by the bin weight calculator; A rectangular weight calculator configured to calculate a weight of the target rectangle based on the bin weight calculated by the bin weight calculator; And A bin weight recalculation unit which re-weights the bins of the color histograms of the object rectangles based on the weights of the color channels and the weights of the object rectangles to extract color features having a strong discriminating power of the object; Color feature extraction apparatus characterized in that it comprises a. 9. The method of claim 8, And the bin minimum distance calculator calculates a minimum distance between the color histogram bin of the target rectangle and the color histogram bin of the non-target rectangle by Equation 12.

(Where J represents one bin of the color histogram of the non-target rectangle of the color channel c, and the bin J represents the bin (i) of the color channel c of the main color histogram of the target rectangle r) Has a minimum distance from and bi is an indication of bin (i).

Is the color histogram of the bin (i) of the color channel (c) of the target rectangle (r),

Is the non-target color histogram of the bins in the color channel (c),

Is the cover of the bin,

Is the minimum distance from the non-target rectangular color histogram bin with the same color channel as the bin (i) of the color channel (c) of the target rectangle (r). 9. The method of claim 8, The bin weight calculator calculates a weight of the color histogram bin of the target rectangle according to Equation (13).

Color feature extraction apparatus, characterized in that for calculating;

(here,

ego,

k is a constant) 9. The method of claim 8, The color channel weight calculator calculates a weight of each bin of the color channel c of the target rectangle r according to Equation 14, and calculates the weight of the color channel of the target rectangle according to Equation 15

Color feature extraction apparatus, characterized in that for calculating;

 Where K represents the number of color channels and M represents the number of bins in the target rectangle histogram. 9. The method of claim 8, The rectangular weight calculator calculates an empty weight of the target rectangular histogram according to Equation 16 (

And calculate the weight of the target rectangle according to Equation 17

Color feature extraction apparatus, characterized in that for calculating;

 (L is the number of target rectangles) The method of claim 12, The bin weight recalculation unit reweights the color histogram bin of the target rectangle according to Equation 18, and reweights the color histogram bin of the target rectangle.

Color feature extraction apparatus, characterized in that to obtain.

The method according to claim 6, And the final color model generating unit generates a final color model of the object according to equation (19).

Where R is the number of target rectangles and M is the number of bins in the target rectangle histogram. The method according to claim 6, And a color blob extraction unit for extracting a color blob of the object from the newly input image based on the final color model of the object. 16. The method of claim 15, And a target position unit which performs a blob analysis on the extracted color blobs, calculates a center and a size of an object of the input image according to the analysis of the blob, and positions and tracks the object. Color feature extraction device. Dividing the target area and the non-target area of the input image into a plurality of rectangles to form a color histogram for each color channel and extracting a main color of the target rectangle from the color histogram; Selecting a bin having a discriminating power from a non-target histogram bin by calculating a minimum distance between the color histogram bin of the target rectangle and the color histogram bin of the non-target rectangle with respect to the main color; Calculating a bin weight of the color histogram of the target rectangle based on the minimum distance; Calculating a color channel weight of the target rectangle and a weight of the target rectangle based on the bin weights; Weighting the color histogram bin of the target rectangle again based on the color channel weight and the weight of the target rectangle to extract color features having a strong discriminating power of the object; And Generating a final color model of the object according to a result of weighting the color histogram bin of the object rectangle again; Color feature extraction method comprising a. 18. The method of claim 17, And calculating a minimum distance between the color histogram bin of the target rectangle and the color histogram bin of the non-target rectangle by Equation 20 in selecting the non-target histogram bin and the bin having the discriminating power.

(Where J represents one bin of the color histogram of the non-target rectangle of the color channel c, and the bin J represents the bin (i) of the color channel c of the main color histogram of the target rectangle r) Has a minimum distance from and bi is an indication of bin (i).

Is the color histogram of the bin (i) of the color channel (c) of the target rectangle (r),

Is the non-target color histogram of the bins in the color channel (c),

Is the cover of the bin,

Is the minimum distance from the non-target rectangular color histogram bin with the same color channel as the bin (i) of the color channel (c) of the target rectangle (r). 18. The method of claim 17, In calculating the bin weight, the weight of the color histogram bin of the target rectangle (

) Is calculated according to Equation 21.

(here,

ego,

k is a constant) 18. The method of claim 17, In calculating the weight of the color channel of the target rectangle and the weight of the target rectangle, the weight of each bin of the color channel (c) of the target rectangle (r) is calculated according to Equation 22, and the target according to (23). The weight of the rectangular color channel (

Color feature extraction method.

 Where K is the number of color channels and M is the number of bins in the target rectangle histogram. 18. The method of claim 17, In calculating the weight of the color channel of the target rectangle and the weight of the target rectangle, an empty weight of the target rectangle histogram according to Eq.

), And the weight of the target rectangle (

Color feature extraction method.

 (L is the number of target rectangles) 22. The method of claim 21, The weight of the color histogram bin of the target rectangle, which is again weighted by re-weighting the color histogram bin of the target rectangle according to Equation 26.

Color feature extraction method characterized in that to obtain.

18. The method of claim 17, And generating a final color model of the object according to equation (27).

(Where R represents the number of target rectangles, M is the number of bins in the target rectangle histogram), R is the number of target rectangles, and M is the number of bins in the target rectangle histogram. 18. The method of claim 17, Extracting a color blob of the object from the newly input image based on the final color model of the object. 25. The method of claim 24, Performing a blob analysis on the extracted color blobs, calculating a center and a size of an object of an input image according to the analysis of the blob, and positioning and tracking the object; Color feature extraction method characterized in that it further comprises.

Images