KR100464079B1 - Face detection and tracking of video communication system - Google Patents

Abstract

The present invention relates to a real-time video communication system based on face extraction and tracking.

The present invention extracts a face region from an input image, encodes and transmits an image including the extracted face region, and at this time, automatically controls a bit rate based on the extracted face region, or extracts the extracted face region. Editing such as compositing with other background areas may be performed to encode and transmit the edited video. The extraction of the face region is performed by detecting a face region for separating a face and a background from an input image, and performing face tracking on subsequent frames based on the detected face region, thereby creating a real-time video communication environment. Implement The face detection is performed by applying a mask for face shape matching based on the skin color, and checking whether the detected face area is a real face through eye, nose, and mouth detection, and tracking the face. Is tracked using a method that limits the position and size of the next face relative to the previous face in successive frames.

Description

Face detection and tracking system in video communication {FACE DETECTION AND TRACKING OF VIDEO COMMUNICATION SYSTEM}

The present invention relates to a system for automatically detecting a face area of a user in communication image data and tracking a face in a continuous frame during video communication.

Recently, a communication method that has been transmitted only by voice has been changed to a video communication of a multimedia environment in which even an image of a communicator is also transmitted. Up to now, such video communication has been mainly used for video communication, video chatting using a PC camera, and video communication using a video phone. However, as the IMT 2000 service is started, a video communication environment using a mobile communication terminal will soon be achieved. Predicting

Video standards such as H.263 and MPEG-4 have already been designated as video standards in mobile video communications for video communications using terminals, and many communication terminal companies are developing terminals capable of high quality video communications.

As such, the communication service using video has become a basic function, but there are still many limitations to satisfy various needs of users for higher quality service. Users' needs can be divided into two types. First, the quality of natural video can be displayed even in a low network environment, and the service aspect that can hide backgrounds that users do not want or reflect their personality. There is this.

In order to solve both of these parts, a technique for separating important objects such as a person from a communication image in real time is required. Such a technology is very difficult at present and has not been achieved in real time. As described above, a conventional technique of separating an object from a given image is as follows.

As a technique of separating objects from a given image, there is a technique of separating partial regions from an image based on a color group. This technique is relatively good at separating areas from an image, but it is a color-based area, not a meaningful object area, and has a long processing time.

Another technique for separating objects from a given image is to separate objects using difference images and edge images. This method enables simple processing of features due to the simple feature information used, but does not separate them from complex screens. However, there is a problem in that the background is stationary and the moving object is separated under the premise that the object is not applicable in the mobile camera environment. In addition, a technique for separating objects by using color information as feature information has been proposed, but this method also requires a significant processing time to use a color after separating a certain area, which makes it difficult to apply in real time and separate moving objects. In the case of the algorithm, for this reason, there is a problem that is difficult to apply to the mobile camera environment.

As such, no algorithm for extracting an object such as a human in real time has been introduced. Instead, few studies have attempted to separate faces by limiting the meaning of objects to faces. In other words, the face separation technique is possible in a relatively real time, but has not yet been useful in applications such as a real-time video communication environment using a mobile terminal in addition to a specific purpose such as a security system.

An object of the present invention is to provide a video communication system that automatically extracts and tracks a face from an image for video communication to enable real-time video communication, a face extraction method, and a face tracking method in the system.

In order to achieve the above object, the video communication system of the present invention includes an input means for image communication for video communication, means for extracting a face region from the input image, and means for encoding an image including the extracted face region. ; Characterized in that comprises a. In another aspect, the present invention is to add an editing effect, such as to synthesize the extracted face region with a different image from the original background, or to encode and transmit the background region separated from the extracted face region at different bit rates. It is done.

In addition, in order to achieve the above object, the face detection method in the video communication system of the present invention comprises the steps of: detecting a face in an image obtained for video communication, checking the face using the detected face as a face candidate area; Encoding and transmitting an image including the identified face area; Characterized in that comprises a.

In order to achieve the above object, the face detection method in the video communication system of the present invention comprises the steps of: extracting a face region candidate from the image of the skin color feature of the face and the elliptical feature of the face; Selecting a face region by detecting eyes, noses, and mouths among the face region candidates; Characterized in that consists of.

In addition, in order to achieve the above object, the face tracking method in the video communication system of the present invention comprises the steps of detecting a face in an input video frame, and using the information of detecting the face of the previous frame in a continuous video communication image. Detecting a face of the frame; Characterized in that comprises a.

1 is a view for explaining a change in image quality according to bit rate control in a video communication system;

2 is a diagram for explaining editing of a face region and a background region.

3 is a flowchart for explaining a face detection method of the present invention.

Figure 4 is a view for explaining the skin color detection in the present invention

5 is a diagram for explaining mask and region matching for face detection in the present invention.

6 illustrates an example of mask and region matching in the present invention.

Figure 7 shows an example of a mask type in the present invention

8 is a view for explaining face matching according to mask type in the present invention.

9 is a view for explaining a face candidate region detection process in the present invention.

FIG. 10 is a diagram illustrating mask matching when only a part of a face appears on a screen in the present invention. FIG.

11 is a view for explaining the influence of the mask size (horizontal / vertical ratio) for real time processing in the present invention.

12 is a view for explaining the necessity of face verification (verification) in the present invention.

Figure 13 is a view for explaining a method for detecting the eye, nose, mouth area for face verification (verification) in the present invention

14 is a view showing an example of detection of the eye, nose, mouth area for face verification (verification) in the present invention

15 is a diagram for explaining the relationship between face initial detection and tracking in the present invention.

16 is a view for explaining a method of limiting a comparison area of a skin color distribution image for face tracking in the present invention;

17 is a diagram showing an embodiment of a video communication system to which the face detection and tracking technique of the present invention is applied.

A video communication system of the present invention for achieving the above object, a face detection method and a face tracking method for video communication will be described as preferred embodiments with reference to the accompanying drawings.

First, in the present invention, a bit rate control and an image editing are described as a background of a technique for implementing a real-time video communication environment.

(1) Bit rate control method of video communication data using face separation technology

In a video communication environment, there is a difficulty in transmitting a large amount of image data through a limited network environment in real time. Since the quality of the image data to be transmitted and the amount of data are proportional to each other, high data rates must be guaranteed at the same time to ensure high image quality. In general, the difference in image quality felt by the user is more severe in the user region than in the entire frame region of the communication image. This is because, in the case of video communication, the main interest of the user and the object to be watched are focused on the user (the call partner).

Therefore, by transmitting only the user region with higher image quality, high image quality can be realized with a low data amount. This feature requires a technology that automatically separates the user from the background.

Examples of bit-rate control and background switching services are shown in Figures 1 (a), (b) and (c). Figure 1 (a) is the original frame image and Figure 1 (b) is a low quality image that can be transmitted when the bit rate is low. However, if the user's face is detected, the quality of the user area and the background area can be adjusted differently. Therefore, as shown in FIG. It is possible to transmit with lower quality by allocating fewer bits.

In the case of (b) and (c) in Fig. 1, the total amount of data finally used is similar, but since the caller's area of interest is that of the other party (mainly a face), the caller has a higher quality of (c) than (b). I feel good. Currently, the H.263 or MPEG-4 based encoding system can adjust the quantization parameter (Quantization parameter) in macroblock units, and thus it is possible to adjust the image quality or data amount described above for each macroblock.

In addition, using the human face extraction technology, if the caller continues to talk in a certain space in the video communication, it is possible to transmit an image of the background space only once, and then only the caller area. In this case, the terminal of the receiver synthesizes and displays the background image originally received and the image of the sender transmitted in real time. The background image transmitted initially may be any background image preferred by the caller, not the actual space where the caller is located. That is, it becomes a kind of video editing function, which is described in the next section.

(2) Image editing of video communication data using face separation technology

This function is a service to edit the communication video to hide the background that the user does not want or reflect the user's personality.

2 shows an example of communicating by switching to a different background such as a background desired by the user or an advertisement background. That is, an example of automatically switching the background to communicate so that the user has an effect such as communicating in another place, and it is also possible to synthesize a specific character or to give a graphic effect as well as changing the background.

In FIG. 2, (a) is a case where the background is switched to another place, (b) is a case where the face region is synthesized with another background image, (c) is a case where the face region is synthesized with a specific character, and (d ) Shows a case in which a face region is synthesized in a frame image, and (e) shows an image in which a smoothing effect is applied only to a portion except for the face region.

In view of the above, the present invention is composed of two types of a face detection method and a face tracking method. Hereinafter, the present invention is divided into a face detection method and a face tracking method.

[Face detection method]

As shown in Fig. 3, the method for detecting a face includes 1. extracting a skin color region, 2. performing an elliptic verification on the skin color region, and 3. checking a face region through eye, nose, and mouth detection. It is made in detail by dividing each process as follows.

[One]. Skin color area extraction

Extraction of the skin color region can be used in a wide variety of methods, the present invention detects the skin color by the following method.

First, when an image is input, a skin color region extraction step of detecting a pixel having a color corresponding to a skin color range in a given image is performed. Human skin color varies widely depending on race, lighting, and device, so it can appear widely. Therefore, the skin color region may include similar color regions in addition to the human region. In order to solve this problem, a skin color grouping step of separating skin color areas which are gathered in a similar color therein is performed.

An example of the result is shown in FIG. In the black-and-white image of FIG. 4, a portion represented by white is an extracted skin color region. However, as shown in (a), (b), and (c) of FIG. 4, the extracted skin color region may include not only a face region but also clothes, a background, and the like, and may not be included in the facial region according to the condition of skin color extraction. Can be.

[2]. Elliptic verification of skin color area

[2.1]. Mask Composition and Matching Value Calculation for Elliptic Matching

As can be seen in FIG. 4, the extracted skin color region alone cannot be determined as a face, and additional means for detecting the face are required. In the present invention, focusing on the feature that the human face shape is elliptical, among the skin color regions extracted in the skin color extraction process, the area whose spatial distribution is close to elliptical is found as a face candidate. Oval features appear on the cheeks and jaw line, even if covered with hair or wearing a hat. In the present invention, an oval mask as shown in FIG. 5 is matched with respect to the skin color region obtained in the skin color extraction process.

The mask of FIG. 5 matches both types of information of the face region and the off-face region. The method is as follows.

It is calculated how many values of the skin color region correspond to when the mask is overlaid on the skin color region image extracted by extracting the skin color region from the skin color region extraction as the face region (Inner Rate). If the values of the white areas of the mask in the skin color image are all identical, the inner rate of the face becomes 100%. The method of matching areas outside the face is similar. The out-of-face area is a black part of the mask and calculates how many mismatched values of the skin color area when the mask is superimposed on the skin color image extracted from the skin color area extraction (Outer Rate). As an example of applying the mask, in FIG. 6, 'Inner Rate' is calculated as 95% and 'Outer Rate' as 98%.

The calculation of the final matching value can be defined according to whether it is important to match the face area and whether it is important to match the out-of-face area, for example to match the face area and to match the out-face area. If important, the final matching value can be defined as:

Matching_value = 0.5 × Inner_Rate + 0.5 × Outer_Rate

Alternatively, if the matching of the face area has twice as important as the matching of the out of face area, the final matching value may be defined as follows.

Matching_value = 0.67 × Inner_Rate + 0.33 × Outer_Rate

For reference, the environment where the matching of the face area is more important than the matching degree of the outside area is when the skin color is not detected well in the face area as shown in FIG. Compared to the degree of matching, an important environment is the case where a lot of skin color is detected in the area outside the face as shown in FIG.

As described above, since the elliptic mask matching method of the present invention calculates the matching degree between the face area and the out of face area separately, a suitable expression for obtaining a matching value according to the application environment is derived by combining the face area matching degree and the out of face area matching degree. There is an advantage to this.

[2.2]. Characteristics of the mask itself

The oval matching mask of the present invention expresses not only the information that the shape of the face region is elliptical, but also the information that the skin color region should not come out of the outside of the face, and in order to eliminate an error occurring in the skin color region of the neck part in the human body, I use a mask without the bottom part.

The following is a comparison of the performance of the mask itself.

As shown in FIG. 7, when there are masks of type A and type B, the result of finding the region having the highest matching value (InnerRate + OuterRate) with the mask of type A in FIG. 7 is as shown in FIG.

As a result of finding the region having the highest matching value (InnerRate + OuterRate) using the B-type mask of FIG. 7, the result of FIG.

In analyzing the results of FIG. 8, the result of the left woman shows that the face type B mask of FIG. 7 reflects the information that the skin color area should not be located above the forehead. In Fig. 7, the mask of type A is compared to the lower end of the ellipse and the face is lowered to the lower part of the neck where the skin color area disappears. Because of the comparison, it can be seen that only the face area is found more accurately.

[2.3]. Oval matching method

Once the oval mask is obtained, the face region is found by matching the oval mask with the skin color image described above. When matching an oval mask to a skin region image, separate a portion of the skin region image into a rectangle, resizing to match the size of the mask, and match the mask to match the inner rate and the out of face region. Calculate the Outer Rate.

Since the position of the separated rectangular area becomes a problem for two variables, a horizontal position and a vertical position on the skin color area image, the rectangular area to be separated is determined by changing the two variables.

In addition, since the size of the rectangular area also becomes a problem for two variables, the length and length are determined, and the variable area for the length and length is changed, and the rectangular area to be separated is determined.

After separating the rectangular regions, the pixels are resized to fit the mask size, and 1: 1 matching is performed with each pixel of the mask, and as a result, matching values (InnerRate and OuterRate) described above are obtained. Finally, a region where the matching value (combination of InnerRate and OuterRate) is greater than or equal to the threshold becomes a face candidate.

This process is illustrated in FIG.

In Fig. 9, PosX is the left starting position of the rectangular region to be separated, PosY is the lower starting position of the rectangular region to be separated, Width is the horizontal length of the rectangular region to be separated, and Height is the vertical length of the rectangular region to be separated.

First, the skin color region image was obtained, and an elliptic matching mask was selected (S11). In the next step (S12), we set 'Rectangle', which is the rectangular area to separate from the skin color image, which is Left = PosX, Right = PosX + Width, Top = PosY, Bottom Set = PosY + Height.

In a next step S13, the rectangular area Rectangle is resized in the skin color image to be equal to the mask size. In a next step S14, the mask matching value Matching_value is calculated by the method described with reference to FIGS. 5 and 6.

In the next step S15, if the matching value is larger than the threshold value recognized as the face candidate, the matching is registered in the face candidate. In the next step S16, the PosX, PosY, Width, and Height values are updated. The updated PosX, PosY, Width, and Height values are checked whether they correspond to the end condition (S17), and if they do not correspond to the end condition, the process continues from step S12. When all executions are finished, face candidates with elliptic matching above the threshold are obtained.

[2.4]. Mask matching when only part of the face appears on the screen

Since the target system of the present invention is a video communication image and the video communication image is not large in image size and the terminal may be moved by the user's movement, the user's face may not be displayed on the screen.

That is, even when only the left side of the user's face is displayed as shown in FIG. 10, the elliptic matching method of the present invention is designed to accurately find the face area in consideration of the hidden part of the face. The area drawn by the ellipse in the left image of the figure is the face region of the user found by the present system.

In order to do this, in the elliptic matching of the present invention, the elliptic matching value of the unhidden portion is set as a Don't care condition, and is represented as a representative value of the entire area including the hidden portion and the unhidden portion. You can use

[2.5]. Adaptive method for real time processing

(One). In the present invention, the resizing step of FIG. 9 is performed a minimum number of times in order to process elliptic matching to find a face candidate in real time.

Resizing is determined for the variables 'Width' and 'Height' for size, so resizing is not possible for rectangular areas with the same 'Width' and 'Height' and different position variables 'PosX' and 'PosY'. It was not repeated. In other words, 'PosX' and 'PosY' are changed, and resizing is performed only once for the entire area referred to, and then the result of resizing of the area referenced for each 'PosX' and 'PosY' value is selected. Is coming.

(2). In order to process the elliptic matching to find the face candidate in real time, the present invention compares the size variables 'Width', 'Height' and the position variables 'PosX', 'PosY' which determine the number of matching to reduce the number of repetitions. Large step size was changed. This makes it possible to increase and decrease the variable value readjusted (updated) in step S16 in FIG. 9, thereby reducing the number of matching iterations.

(3). In order to process the elliptic matching to find the face candidate in real time, the present invention removes the height (Height) from the four variables (Width, Height, PosX, PosY) for the size and position, and the width. We used a pre-defined vertical length for (e.g., when the length is set to '1', the length is set to '1.2', or when the length is set to '1', the length is set to '1.3'). In this case, there is a disadvantage in that the ratio of the length of the face to the length of the face does not correspond to the face, but as shown in FIG. 11, the result shows that the difference is not significant in detecting the face.

That is, in FIG. 11, the first person and the third person have a more accurate face boundary when the ratio of the face is 1: 1.2, and the second person has a more accurate case where the ratio is 1: 1.3. While expressing face boundaries, the difference in results is not significant in the sense of detecting a face.

(4). In order to process the elliptic matching to find a face candidate in real time, the present invention uses a 32 × 33 size elliptical mask that is much smaller than the face size of a person appearing in a normal video communication image. In this way, the face candidate area is searched with the small size of the elliptical mask, so the process of finding the candidate is faster and faster than using the large size of the elliptical mask.

[3]. Facial region identification through eye, nose and mouth detection

After finding an elliptical region that is a candidate for the face by the above-described method, it is necessary to detect eyes, nose, and mouth to check whether the elliptical skin color region is actually a face. For example, as shown in FIG. 12, since the elliptic matching value of not only the face but also the hand is high in the skin color distribution image, it is extracted as the face candidate region. In this case, a clue that can distinguish the real face from the non-face part may be the presence of eyes, nose and mouth.

For this purpose, the method of detecting the eyes, nose, and mouth consists of two steps, the first step being to find the dark and upper and lower part of the light in the elliptical area, and the second step to find the actual eye from the first step. It consists of determining the nose, mouth.

(One). How to find the dark part of the center and the bright part of the top and bottom of the ellipse area

In order to find the dark part of the elliptic area, the basic pattern as shown in FIG. 13 (a) is matched with a gray scale image for the original image. The basic pattern is that the middle band is dark and the upper and lower bands are bright because the eyes, nose, and mouth of the face are surrounded by skin areas. And because the brightness is above and below, you can get rid of dark areas like the hair area. The size of the pattern shown in Fig. 13A is designated as a value proportional to the width of the ellipse found in the elliptic matching.

Figures 13 (b) and 13 (c) are results of finding dark portions of the face using the pattern of (a). As shown in the black and white skin color distribution image below the original image, the area indicated by the square around the eyes, nose and mouth is detected as the dark part of the face. As shown in the results, the eyes, eyebrows, nose, mouth and lower chin were detected as dark areas of the face.

(2). Steps to determine the real eye, nose and mouth of the dark

As described above, the method of determining the eyes, nose, and mouth in the dark areas of the face found using the basic pattern uses a spatial relationship in which the eyes, nose, and mouth may be located. That is, the eyes are at the top of the face, both eyes are at a similar height, the nose is at the bottom between the two eyes, and the mouth is below the nose.

Figure 14 shows an example of the results finally selected from the dark areas of the face with eyes and mouth.

[Face tracking method]

Face tracking is a method of real time face detection at the time of video communication by detecting the face area at a faster time after the initial detection of the face by the above-described method and subsequent video communication frames.

The relationship between the face initial detection and the face tracking is as shown in Fig. 15, the initial face is detected (S21), the detected face area is examined to satisfy the initial face condition (S22), and if the initial face condition is not satisfied, the next frame is detected. Detects the initial face with respect to the face, and performs face tracking if the initial face condition is satisfied (S23), examines whether the tracked face satisfies the face tracking condition (S24), and if the face tracking condition is not satisfied, the initial face. If detection is performed and the face tracking condition is satisfied, face tracking is continued for the next frame.

That is, the initial face region is found using the first frame (or several consecutive frames to increase the accuracy) (S21, S22), and then the face tracking method is applied to the next frames (S23, S24). Since the face must be detected in real time during video communication, the face tracking method is simpler but faster than the initial face detection method. In addition, when an error occurs in the face tracking step or periodically, it may be reinitialized using an initial face detection method.

The face tracking method is basically the same as the initial face detection method, but reduces the number of comparisons of elliptic matching. There are two ways to reduce the number of comparisons of elliptic matching. The first method is to limit the comparison area of the skin color distribution image, and the second method is to limit the degree of change in face size.

(One). How to limit the comparison area of skin color distribution image

In the initial face detection, when the elliptic matching is performed, the entire area where the skin color exists in the skin color distribution image is searched, but in the face tracking step, the part of the face centered on the area where the face of the previous frame is located Search only). For example, in Fig. 16, in the initial face detection step, all skin color present areas (outer square area) such as face and neck, and the vest area belonging to the skin color are searched, but in the face tracking step, as shown in FIG. Only the marked part will be searched.

(2). How to limit the gradient of face size

In the face tracking step, the size of the ellipse to be compared is limited to the range specified for the previous face size. For example, the size of the ellipse S (t) possible in the current frame is for the ellipse size S (t-1) of the previous frame;

0.9 x S (t-1)? S (t)? 1.1 x S (t-1).

For reference, in the initial face detection step, threshold values for the maximum face size and the minimum face size in the video communication image were previously determined, and then all possible face sizes in the minimum / maximum threshold range were compared.

[Image communication system]

17 shows an example of the configuration of a video communication system to which the face detection and tracking method described so far is applied. The image input unit 100 obtains a user image for image communication, and may be, for example, a PC camera. Face region extraction means 101 extracts the face region from the image obtained by said image input means (100). As described above, the face region extracting means 101 includes a face detecting means 101a for separating and detecting a background and an object (face) from an input image, and a face tracking means for performing face tracking on the detected face. 101b may be included. The video signal including the face region (or the face region to be tracked) extracted by the face region extraction means 101 and its background is encoded and transmitted by the encoding means 102. The encoding means 102 may include bit rate control means 102a for automatically controlling a bit rate based on the extracted face region. The bit rate control unit 102a encodes the face region at a high bit rate and transmits the background region at a lower bit rate, as described with reference to FIG. The less-interested background can be processed at lower quality, providing a video call environment that can be perceived with good image quality while reducing the overall transmission burden.

In addition, the editing means 103 performs image editing as described with reference to FIG. 2 based on the extracted face region. For editing an image, the editing means 103 may have an interface with a user, and may have a memory in which information such as a background image for various editing effects is stored in advance.

The video communication system configuration of the present invention shown in FIG. 17 can be variously modified and implemented based on the video input means, face extraction means, encoding means, and editing means, and is not limited to the configuration shown in FIG. Although not shown in FIG. 17, since the video communication system performs bidirectional transmission of audio and video, it is obvious that an element for decoding and displaying the video signal transmitted from the counterpart is added.

The present invention provides a system that can extract the face region from the video communication data in real time through face detection and tracking technology, and effectively control the bit rate of the video communication or automatically edit the video communication image using the extracted face region information. do.

The present invention extracts the caller's face area, which is the area of interest of the user, in the video communication system, and transmits the face area in high quality compared to the background area based on this. The background image can be edited as necessary to provide various additional functions in video communication.

Claims (19)

Image input means for inputting an image for video communication; The face region is extracted from the initial frame of the image input through the image input means, and in the subsequent frames, the face region is obtained by elliptic verification of the skin color region with respect to the current frame based on the face region position extracted from the previous frame. In extracting a face, face region extraction means for extracting a face region while changing the degree of change in the size and position of the face of the previous frame only within a specified range; Encoding means for encoding an image including the face region extracted by the face region extracting means ; Rate control means for automatically controlling the bit rate based on the face region extracted by the face region extracting means; And a video communication system comprising a. The video communication system according to claim 1, further comprising editing means for editing an image based on the extracted face region. delete In an image acquired for video communication, an elliptical face region represented by a parametric expression is detected in an initial frame of the acquired image, and in a subsequent frame, a face region of the elliptical face is detected based on the position of the face region detected in the previous frame. In detecting a face region through elliptic verification of the skin color region, detecting a face region while changing the degree of change in the size and position of the face of the previous frame only within a specified range , Determining whether a face is a face candidate area, encoding and transmitting an image including the identified face area; Face detection method in a video communication system comprising a. The method of claim 4, wherein the detecting and confirming of the face comprises: extracting a skin color region, performing a facial shape verification by comparing an elliptical degree of the skin color region using a mask in the extracted skin color distribution image; Identifying a face region through eye, nose, and mouth detection with respect to the facial shape verified result; And a face detection method in a video communication system. 6. The face type verification using the mask according to claim 5, wherein the facial shape verification using the mask uses two pieces of information that the skin color distribution of the face area is close to an oval, and that the outside area has no skin color distribution. Detection method. The method of claim 5, wherein the performing of the face shape verification comprises quantifying a combination of two pieces of information indicating that the skin color distribution of the face region is close to an oval, and that the information outside the face has no skin color distribution. A face detection method in a video communication system, characterized in that the face candidate is selected as a face candidate when the value of the ellipticity is digitized. 6. The method of claim 5, wherein comparing the elliptical degree of the skin color area using a mask in the extracted skin color distribution image comprises: separating a portion by adjusting a position and size larger than one pixel unit in the skin color area; Resizing a part to be equal to a mask size that is smaller than the actual face size, and comparing the result with respect to each pixel for the mask, thereby detecting a face. Way. The method of claim 5, wherein the face shape verification for the skin color region is regarded as a do n't care condition when only a part of the caller's face appears on the screen, and the ellipticity of the displayed part is digitized. A face detection method in a video communication system, characterized by using a representative value for the entire area where a part showing one value and a part not appearing are used. The method of claim 8, wherein the resizing is performed only once for the entire area resizing at the same size ratio for faster calculation. delete delete Detecting a face region in an initial frame of an input image, and extracting a face region through elliptic verification of a skin color region with respect to a current frame based on the position of a face region detected in a previous frame in a subsequent frame. Extracting a face region while changing the degree of change in face size and position of the frame within a specified range ; Face tracking method in a video communication system comprising a. delete The method of claim 13, wherein the limiting of the degree of change while changing only within a specified range is to adjust the size or position only within a range specified for the face size or position of a previous frame. A face extraction step of extracting a face region by detecting a skin color region and an elliptic mask matching in an input image, based on the extracted face region, and in a subsequent frame, based on the position of the face region extracted from the previous frame In detecting a face region through elliptic verification of the skin color region for a frame, a face tracking step of detecting a face region while changing the degree of change in the size and position of the face of the previous frame only within a specified range , and detecting the detected face region and the background. Separating and transmitting the face region and the background at different bit rates; Encoding control method of a video communication system comprising a. The method of claim 16, wherein the eye, nose, and mouth are detected by matching a predetermined reference pattern for the detection of eyes, nose, and mouth with respect to the face area within the detected face area, and the face area is detected from their spatial positional relationship. Encoding control method of a video communication system, characterized in that for verifying. delete 17. The method of claim 16, wherein a video editing effect is provided by synthesizing and encoding the separated face region with a new background image replacing the original background region.