KR100286742B1 - Method of detecting scene change and article from compressed news video image - Google Patents

Abstract

The present invention relates to a method for efficiently indexing compressed sports news videos, and more particularly to a method for detecting scene transitions and news articles in a compressed domain.

The present invention provides a first step of detecting a scene transition possibility by applying a brightness difference ΔY and a color difference correlation ρ _n between an nth I-frame and an n + 1th I-frame to a pseudo neural network. In all P-frames included in the nth GOP (Group of Picture) in which the possibility of transition is detected, the number of intra macroblocks is divided by the number of forward prediction macroblocks to determine the P-frame constant ( In the second step of obtaining R _p ), in all the B-frames included in the n-th geophysics for which the scene change possibility is detected, the number of backward predicted macroblocks is divided by the number of forward predicted macroblocks to determine the B-frames necessary for the scene change detection. the constant (R _b) to obtain the fourth step of detecting a correct scene change frame in the step 3, using the constant frame P- and B- frames constant the scene change is detected jiohpi possibility, and the fourth step Standing applied to the scene change frame accurate editing special effect detection algorithm is not detected, and provides a transition of the compressed video news detecting method, characterized in that, including a fifth step of detecting a specific operation of the camera frame.

Description

{Method of detecting scene change and article from compressed news video image}

The present invention relates to a method for efficiently indexing compressed sports news videos, and more particularly to a method for detecting scene transitions and news articles in a compressed domain.

Recently, with the rapid increase in the use of digital video, the development of a multimedia service system capable of providing various functions by video indexing has been made. One of the technologies required for such a multimedia service is an algorithm capable of detecting various scene changes (cut, dissolve, wipe). Here, cut refers to a case where a scene changes abruptly, and dissolve and wipe refer to a case where a scene is gradually changed by inserting a special editing effect between scene change points.

On the other hand, because digital video requires a lot of storage space, compression of these video data is essential for storage and transmission. As such compression standards, MPEG-1, MPEG-2, or Motion JPEG are widely used. Today, the object-oriented encoding standard of MPEG-4 has been completed.

Algorithms used to detect scene transitions in compressed video such as MPEG or Motion JPEG can be divided into two types. One method is to detect scene transitions while minimizing decoding of compressed image data. Another method is to detect the scene change after decoding the compressed image data.

The latter method can detect the exact scene change point, but has a disadvantage of slow processing because it undergoes a decoding process such as motion compensation or Inverse Discrete Cosine Trasform (IDCT). On the other hand, the former method has been recognized in terms of accuracy and accuracy in the process of detecting the scene change, the scene change detection method using the former method has become a global trend.

As described above, a conventional method of detecting a scene change of an image in the compressed domain is as follows. The first method is a paper published by Yasuyuki Nakajima, Kiyono Ujihara, Akio Yoneyama, et al., 'Universal Scene Change Detection on MPEG-Coded Data Domain', SPIE Vol. 2419, 1996. This method obtains a DC image consisting of DCT DC coefficients in an I-frame, obtains a DC image through motion compensation in a P-frame and a B-frame, and then compares the reconstructed DC images to detect a scene change. It is a way. Since this method does not go through the inverse discrete cosine transform (IDCT) process, the processing speed is improved.

The second method is published in Jianhao Meng, Yujen Juan, Shin-Fu Chang et al., 'Scene Change Detection in a MPEG Compressed Video Sequence', SPIE Vol. 3024, 1996. This method uses macroblocks (MB) types of P-frames and B-frames, which can reduce processing time even more because they do not undergo motion compensation and IDCT processes. However, this method not only detects transitions in I-frames when the GOP structure is closed GOP but also detects progressive transitions such as dissolves and wipes. There is a disadvantage.

The third method is a paper presented by Na-Yeop Woo, Seok-Hoon Woo, Chi-Won Won, and Hee-Il Han, 'Detection of Scene Change by Measuring Data Generation in MPEG-2 Bitstream', Signal Processing Conference, Vol. 10, pp.69-72 , 1997. This method detects the scene change point by using the amount of change in the number of edges between I-frames and assumes that the edges between blocks exist when the difference of DC coefficients between 8x8 blocks changes rapidly. This is a method for detecting a scene change.

As described above, many algorithms for detecting a scene change in MPEG-based compressed image data have been developed. However, conventional algorithms accurately detect the scene change point of a fast-moving image such as sports news and a video having a lot of change in lighting. There was a problem that could not be detected.

Accordingly, the present invention has been made to solve the above problems of the prior art, an efficient scene for video such as sports news with a lot of relatively fast movements such as sports news and a lot of changes in lighting, etc. Its purpose is to provide a conversion detection method.

Also, scenes such as cuts and progressively changing dissolves or wipes, such as a sudden change in the screen, using only the information contained in the compressed video stream without decoding the video compressed with MPEG. Another object is to provide a method for detecting a switch point.

Another object of the present invention is to provide a method for detecting an anchor frame in a compressed domain and detecting a news article between the anchor frame and the anchor frame.

1 is a flowchart illustrating a scene change detection method of a compressed news image according to a first embodiment of the present invention;

2 is a structural diagram of a pseudo neural network for detecting a scene change possibility between I-frames according to the first embodiment of the present invention;

3 is a structural diagram of a closed geoppy shown to illustrate the present invention;

4 is a structural diagram of a non-closed geopi shown to explain the present invention,

FIG. 5 is a graph showing a relationship between time and the number of intra macroblocks of a P-frame according to a type of special editing effect of a camera;

6 is a graph for detecting a special editable effect insertion section;

7 is a view illustrating a method of detecting a wipe among special editing effects;

FIG. 8 is a view illustrating an anchor frame without a news icon and an anchor frame with a news icon in order to explain a second embodiment of the present invention.

In order to achieve the above object, a first embodiment of the present invention applies a brightness difference ΔY and a color difference correlation ρ _n between an nth I-frame and an n + 1th I-frame to a pseudo neural network. In the first step of detecting a scene change possibility, and in all P-frames included in the nth GOP (Group of Picture) in which the scene change possibility is detected, the number of intra macroblocks is determined as the number of forward predicted macroblocks. In the second step of dividing to obtain the P-frame constant (R _p ) necessary for the scene change detection, in all B-frames included in the n-th geophysics where the scene change possibility is detected, the number of backward predicted macroblocks is determined by the number of forward predicted macroblocks. accurate scene change frame in a frame divided by the number of B- constant required for the scene change detection (R _b) to obtain a first step, by using the constant frame P- and B- frames constant the transition potential is detected jiohpi A fourth step of detecting, and a fifth step of detecting a special manipulation frame of the camera by applying the special editing effect detection algorithm if the correct scene change frame is not detected in the fourth step. A scene change detection method is provided.

In addition, the second embodiment of the present invention, in the method for detecting the beginning and end of the TV news article interposed between the image of the anchor (anchor frame), the position of the anchor face according to the type of the anchor frame In step 1, if the face color is greater than or equal to a threshold value, a first step of setting a candidate anchor frame, a second step of obtaining DCT energy distributions of DC coefficients and AC coefficients of DCT blocks constituting the candidate anchor frame, and the candidate anchor frame A third step of obtaining averages and variances of DCT energy distribution of each block by dividing into face and background parts, and averaging and variance of DCT energy distribution of the candidate anchor frame by average and variance of DCT energy distribution of anchor frame A fourth step of measuring similarity between the candidate anchor frame and the anchor frame by comparing with each other, and the candidate anchor according to the similarity Providing a fifth step, the article detection method of compressed video news, characterized in that, including a sixth step of detecting the anchor frames between the article area to set the frame as the anchor frame.

Hereinafter, exemplary embodiments of the present invention will be described with reference to the accompanying drawings.

[First Embodiment]

1 is a flowchart illustrating a scene change detection method of a compressed news image according to a first embodiment of the present invention. The present invention detects a scene change point of a compressed news image using a hierarchical cut detection algorithm.

In general, a video sequence compressed with Moving Picture Experts Group (MPEG) has a GOP (Group of Pictures) structure, and each GOP necessarily includes one intra-frame. One GOP is composed of 15 frames of 0.5 seconds and includes several P-frames (forward-predictive coded video frames) and B-frames (bi-predictive coded video frames). Therefore, in the scene change search, only the I-frames of each GOP are compared to see if the scene change is made in the P-frame and the B-frame therebetween only when there is a sudden scene change between them.

As such, the scene transition detection process between I-frames is as follows.

MPEG-compressed video uses the Y-Cb-Cr color coordinate system. Where Y represents a brightness component and Cb.Cr represents a color difference component. The difference in brightness of the pixels (difference of Y component: DELTA Y) is also very large due to the large movement of the object or the special operation of the camera (for example, fan and tilt). Therefore, when only the brightness difference (ΔY) information is used, there is a case where the camera is misidentified as a scene change even by special operation of the camera.

On the other hand, since the color difference correlation due to the Cb · Cr component is relatively insensitive to the large movement of the object and the operation of the camera, it is possible to detect whether the scene change is more accurate when the brightness difference and the color difference correlation of the pixel are used together.

The above-described principle is used to detect a transition between I-frames. First, when a GOP of an MPEG compressed video image is input (S11), I-frames are used by using DCT coefficients of I-frames. The DC image of the frame is obtained, and the DC image of the P-frame and the B-frame is obtained using the DC image of the I-frame and the motion compensation value (S12). Next, the brightness difference and the color difference correlation of the DC image between the I-frames are calculated (S13). Here, the brightness difference between the I-frames is obtained as in Equation 1, and the color difference correlation between the I-frames is obtained as in Equation 2.

Here, Y _n is a histogram of the DC component of the Y component in the current I-frame, Y _n-1 is a histogram of the DC component of the Y component in the previous I-frame.

Where H _n is the histogram of the Cb and Cr components in the current I-frame, and H _n-1 is the histogram of the Cb and Cr components in the previous I-frame.

When the brightness difference ΔY and the color difference correlation ρ _n between the I-frames are obtained as described above, it is applied to the pseudo neural network (S14) to detect whether the scene is changed (S15). That is, when there are many fast movements such as sports news or camera manipulations (up / down, left / right movement, zooming in / out of the image), it is difficult to set a threshold value for scene change. To solve this problem, we use pseudo-neural networks. The pseudo-neural network receives the brightness difference ΔY and the color difference correlation ρ _n between the I-frames and determines whether or not there is a scene change.

This pseudo neural network is shown in FIG. 2, and the two input layers 21 receive the brightness difference ΔY and the color difference correlation ρ _n between the I-frames, respectively, which are hidden layers. Two output values are output through two output layers 23 via a hidden layer 22. Based on this value, it is determined whether or not there is a transition. By using the neural network as described above, not only the threshold value can be determined using the training data, but also the threshold plane is set non-linearly to prevent misrecognition of the scene change detection.

When the scene change possibility is detected between the two GOPs through the above process (S15), the process proceeds to the next step to detect the scene change point (S16), and the possibility of the scene change between the two GOPs. If not detected, check the next geopi.

When the possibility of a transition is detected between I-frames included in two geopis, the transition point is limited to one geopi. There are two ways to find the transition frame within the geopi, depending on the geopi's structure. . That is, the geophysical structure is divided into closed and non-closed geophysics according to whether or not the images in the geophysics can be independently reproduced from other geophysics. The scene change detection algorithm according to the geophysical structure will be described later.

In order to detect an accurate scene change point, P-frame constants and B-frame constants of P-frames and B-frames constituting the detected geometries are obtained, respectively.

I-frames can be decoded by themselves without reference pictures because the entire picture is encoded as intra macroblocks (MB _intra ), but macroblocks in P- and B-frames are encoded using motion vectors. Depending on the type of the macroblock, there may be decoding only if there is a past or future reference picture.

If there is a lot of motion in the scene change frame or image, there is no part to refer to from the past image, so the number of intra macroblocks (MB _intra ) in the P-frame increases and the number of forward predicted macroblocks (MB _forward ) is increased. Decreases. In addition, the number of _forward macroblocks (MB _forward ) in the B-frame is reduced while the number of macroblocks (MB _backward ) in the reverse direction increases. Scene change frame is detected using the same principle.

Presence or absence of a scene change in every other frame is the scene change possibility excluding I- frames within the detected jiohpi (GOP) is there is detected by the scene change detection algorithm, using the equation (3) and a P- frame, the constant (R _p) B Find the frame constant (R _b ).

If there is a scene change or there is a lot of motion, the value of the P-frame constant (R _p ) increases and the value of the B-frame constant (R _b ) has a very large value.

In the above, there are two types of geophysical structures, closed and non-closed, and it is mentioned that a separate scene change detection algorithm is required according to the geophysical structure.

First, an algorithm for detecting a scene change frame in the case of a closed geoppy will be described with reference to FIG. 3.

Note that when detecting scene change points in MPEG-compressed video data, the order in which the frames are output on the screen and the encoding order are different. A geop with a closed geop structure with two B-frames between the P-frames is shown in FIG. 3.

In the closed geophysical structure, the scene change frame detection algorithm is as follows.

First, the presence or absence of a scene change is checked in sections ①, ②, and ③ of the drawing except for the I-frame in the geofi. ① If the Nth P-frame constant (R _{p_n} ) is larger than the threshold value (T ₁ ) in the section, ① It is determined that there is a possibility of scene change in the section, but because there may be a scene change in the B-frame preceding in time, -It is determined whether or not the scene transition in the frame.

That is, if the N + 1th B-frame constant R _{b_n + 1} is greater than the threshold value T ₂ , it is determined that there is a scene change in the N + 1th B-frame. However, if the N + 1 th B-frame constant R _{b_n + 1} is not greater than the threshold T ₂ , and the N + 2 th B-frame constant R _{b_n + 2} is greater than the threshold T ₂ . It is determined that there is a scene change in the N + 2th B-frame. If _{neither the} N + 1th B-frame constant (R _{b_n + 1} ) nor the N + _2th B-frame constant (R _{b_n + 2} ) is greater than the threshold (T ₂ ), then the transition in the Nth P-frame It is determined that there is. Of course, if the Nth P-frame constant R _{p_n} is not larger than the threshold value T _{1 in the} section ( ₁ ), it is determined that there is no scene change in the section ( ₁ ), and it is determined whether the scene is changed in the next section.

If the scene change frame is not detected in all sections (①, ②, ③) through the above process, it is finally checked whether the scene change occurs in the I-frame of the next geop, that is, the N + 9th I-frame. That is, if the number of intra macroblocks MB _intra in all P-frames of the current geopy is less than or equal to the threshold, it is determined that a scene change occurs in the N + 9th I-frames. When the correct scene change frame is detected through the above process (S17), the cut scene is detected by storing the corresponding scene change frame (S18). However, if the number of intra macroblocks in all the P-frames is not less than or equal to the threshold value, the special manipulation frame of the camera is detected by applying to the special editing effect detection algorithm (S19).

Next, an algorithm for detecting a scene change frame in the non-closed geophysical structure will be described with reference to FIG. 4.

In the case where the geophysical structure is non-closed geophysical, it is first determined whether or not the scene transition between the I-frames is possible, and then whether the scene transition in the P-frame and the B-frame in the potential geop is detected. In the case of non-closed geo-fiber, the B-frames immediately following the I-frame are past images in time rather than the I-frame. Therefore, in order to find the correct transition frame in the non-closed geocode with the possibility of transition, the section ② and ③ composed of P-frames and B-frames are first examined, and if the transition frame is not detected, the ④ of the next geo-fi Examine the section. Here, the B-frame (section 1) following the N-1th I-frame of the current geophy need not be examined because it is a past image in time than the N-1th I-frame.

The scene change frame detection algorithm in section 2 and section 3 is the same as that of closed geopi. When a scene change frame is detected in section (2) or section (3), it detects whether the scene is changed in the next geo. However, if the transition frame is not detected in sections ② and ③, it checks whether the number of intra macroblocks (MB _intra ) of all the P-frames in the geo is less than or equal to the set threshold value. do.

However, if the number of intra macroblocks (MB _intra ) of the P-frames in the geop is less than or equal to the set threshold value, the special editing effect detection algorithm is used to determine whether the special editing effect is inserted.

(4) The algorithm to find the transition frame in the section is as follows.

First, when the frame constant R _{b_n + 9 of the} N + 9 th B-frame is greater than the threshold value T ₁ , the N + 9 th B-frame is detected as the scene change frame. The frame constant (R _{b_n + 9} ) of the N + 9th B-frame is not greater than the threshold (T ₁ ), and the frame constant (R _{b_n + 10} ) of the N + 10th B-frame is the threshold (T ₁ ). If large, the N + 10th B-frame is detected as a scene change frame. However, if the N + 9-th frame of constant B- frame (R _{b_n + 9)} and the frame N + constant of the 10th frame B- (R _{b_n + 10)} are both greater than the threshold value (T ₁₎ N + 8 beonjjae Detects an I-frame as a transition frame.

The scene change detection algorithm described above can detect radical scene changes (cut scenes) but cannot detect progressive scene changes (dissolves, wipes, fade ins / outs, etc.). Therefore, an additional algorithm for accurately detecting the portion where the special editing effects are inserted is required, which is called a special editing effect detection algorithm.

The duration of the special editing effects (dissolve, wipe, fade in / out, etc.) of the camera is about 1 second, and using the change state of the number of intra macroblocks of P-frames within two geometries, Can be detected. 5 is a graph illustrating a relationship between time and the number of intra macroblocks of a P-frame. As the dissolve proceeds, the number of intra macroblocks of the P-frame changes as shown in (a), and as the wipe progresses, the number of intra macroblocks of the P-frame is (b). Changes as shown in

The comparison of the dissolve and wipe characteristic curves shows that the intra macroblock (MB _intra ) from the start frame of the P-frame immediately before the special editing effect to the P-end frame immediately after the special editing is finished. ) Is used. As shown in FIG. 7, the number of intra macroblocks of the P-frame in the special editable effect insertion section, that is, the search region, is larger than the set threshold value, and the slope of the number of change increases rapidly and then decreases rapidly. Therefore, when the number of intra macroblocks of the P-frame in the MPEG-compressed video data is shown in FIG. However, if the number of intra macroblocks of the P-frame does not appear as shown in the graph of FIG. 5, the next geopee is examined without recognizing it as a transition point as a part of camera operation or a large object movement.

Once the special editable insertable interval is set, DC images are obtained for the post-finished images from the image before the special edit effect occurs. From the DC images thus obtained, it is determined whether the special editing effect is a dissolve or a wipe.

This judgment uses the image that changes as the dissolve and wipe proceed. In the case of dissolve, as the time progresses, the overall change of the image occurs, and the images appearing when the dissolve is started gradually become closer to the image after the dissolve is completed. In the case of wipes, on the other hand, the change starts from a part and spreads out to the whole image.

The dissolve can be obtained using only the variance of the difference between DC images, but since the wipe is difficult to distinguish from the images shown by the operation of the camera (Panning, Tilting), the wipe is performed by partially labeling the image as shown in FIG. Detect.

In general, in the case of dissolve, when the images changed after dissolving are compared with each other, the change occurs in the whole image (both ① to ⑤). Since the dissolve is not a sudden change of the scene, when the images changed after dissolving are compared with each other, the amount of change in each labeled area is smaller than the set threshold value. When this phenomenon occurs, it is determined that the dissolve is inserted.

On the other hand, in the case of a wipe, since the change of the image starts from a part (one of ① to ⑤) and spreads out to the whole image, the wipe can be detected by comparing the images changed after the wipe starts. That is, when the image before the wipe and the image where the wipe is started are compared with each other, if the number of changed regions gradually increases, it is determined as a wipe.

However, as a result of examining the insertable section of the special editing effect, if it is not a dissolve or a wipe, it is a case where a camera movement (tilting, panning, zoom in / out, etc.) or a lot of movement is included in the image. Since the proposed scene change detection algorithm does not recognize this part as a scene change, misrecognition can be reduced.

Second Embodiment

A second embodiment of the present invention is a method for detecting articles in a compressed news image.

TV news consists of several articles, and one article consists of several shots. Therefore, the most important thing for the index of TV news is to accurately detect the start and end points of articles. Due to the nature of TV news, the article starts with the anchor (the anchor frame) and the content comes out, and when the anchor comes out again, another article begins. This feature can accurately detect the multiple articles that make up TV news. In other words, if you find the anchor frame correctly, you can find the beginning and end of the article.

In TV sports news, only one anchor is present, the anchor is fixed, and there is little movement in the video except for the small movement of the anchor. This feature is used to designate the part where the face of the anchor appears, and to detect the anchor frame using the motion vector value. That is, the motion vector value is very small in the anchor frame.

The anchor frame can be divided into a case where there is no news icon as shown in (a) and a case where there is a news icon as shown in (b), and the position of the anchor face and the news icon without a news icon are shown. Since the position of the anchor face in the case is different from each other, specify different positions. The detection of the anchor frame uses the I-frame, and in the case of the anchor candidate frame, the motion vector of the P-frame following the I-frame is examined and finally determined as the anchor frame if there is almost no motion.

The detailed process for this is as follows.

As shown in FIG. 8, when the face color is greater than or equal to a threshold value at the position of the designated anchor face (1: when there is no news icon, 2: when there is a news icon), it is set as a candidate anchor frame that may be an anchor frame. . However, in this case, when the face of the person other than the anchor is positioned at the designated position according to the image, there is a case in which it is misidentified as the anchor frame. To compensate for this, CCV (Color Coherent Vector) is used.

That is, the anchor frame is divided into the anchored part and the background part according to the type, and the energy distribution of the DCT is obtained by using Equation 4 in the 8 × 8 blocks included in each part.

First, the DCT energy distributions of DC coefficients and AC coefficients are first obtained for 8 × 8 blocks, and the average and variances of DCT energy distributions of the blocks obtained from the face part and the background part are obtained, respectively. Used as a texture to detect.

That is, the DCT energy distributions of the DC coefficients and the AC coefficients of the DCT block are obtained with respect to the candidate anchor frame, and the average and the variance of the DCT energy distributions of the blocks obtained from the face part and the background part are respectively obtained, and then the anchor used as a texture. The similarity is measured by comparing the average and the variance of the frame. Since video data compressed by MPEG is compressed on the basis of DCT, the DCT energy distribution is used to measure the similarity with the anchor frame, so that the feature vector having excellent discrimination can be extracted from the compressed region without the inverse DCT process.

As such, when the anchor frame is detected, the start point and the end point of the TV news article interposed between the anchor frames can be accurately detected.

KBS Sports News with 704 × 480 resolution encoded in MPEG-2 was used for scene change detection experiments including special editing effects (dissolve, wipe, etc.). Here, one GOP is composed of 19 frames, and the video data of which the GOP structure is a closed GOP is used. The article detection results obtained through the proposed algorithm for KBS Sports News are shown in Table 1, and the scene change detection results are shown in Table 2. According to the experimental results, the article was 100% accurate and the detection result was more than 97% accurate. In detecting special editing effects, the recognition rate was slightly lower (75% dissolve, 77,5% wipe), but the special editing effect detection algorithm could not detect all the editing effects because the special editing effects were so diverse.

Kinds KBS Sports News Detection result accuracy(%) Article 7 7 100

Kinds KBS Sports News Detection result accuracy(%) Cut 185 180 97.2 Dissolve 8 6 75.0 Wipe 9 7 77.7 Sum 202 192 95.0

As described above, according to the present invention, a scene change point can be efficiently detected for a video having a lot of relatively fast movements such as sports news and a lot of changes in lighting, and so on without being decoded the compressed video. Since the scene change point can be detected, it is possible to save time according to decoding and accurately detect the scene change point in a short time.

In addition, since it is possible to detect anchor frames in the compression domain and to detect news articles interposed between the anchor frames, it is possible to index compressed news videos, thereby providing various multimedia services.

Claims (5)

a first step of detecting a scene transition possibility by applying a brightness difference ΔY and a color difference correlation ρ _n between an nth I-frame and an n + 1th I-frame to a pseudo neural network; In all P-frames included in the nth group of pictures (GOPs) where the possibility of transition is detected, the number of intra macroblocks is divided by the number of forward prediction macroblocks to determine the P-frame constant (R). the second step to find _p ), In all B-frames included in the n-th geophysical detectable scene transition, a third frame for obtaining a B-frame constant (R _b ) necessary for scene change detection by dividing the number of backward prediction macroblocks by the number of forward prediction macroblocks. step, Using the P-frame constant and the B-frame constant, detecting a correct scene change frame in the geo-fiber where the scene change possibility is detected; and And a fifth step of detecting a special manipulation frame of the camera by applying to a special editing effect detection algorithm if the correct scene change frame is not detected in the fourth step. The method of claim 1, wherein the fourth step, A first sub-step of combining one encoded P-frame into one section by referring to the P-frame and the P-frame as a future image when the geophysical structure is a closed geometry; A second sub-step of detecting the B-frame as a scene change frame when the P-frame constant is greater than a threshold and the B-frame constant is greater than a threshold in the bundled section; A third sub step of detecting the P-frame as a scene change frame when the P-frame constant is greater than a threshold and all of the B-frame constants are less than a threshold in the bundled section; A fourth sub step of determining that there is no scene change in the corresponding section when the P-frame constant is smaller than a threshold value in the bundled section, and performing the second and third sub-steps in all the bundled sections of the current geopie. , A fifth sub-step of comparing the number of intra macroblocks in all P-frames in the current geopi with a threshold when no scene transition frame is detected in all bundled sections of the current geopi, A sixth sub-step of detecting an I-frame of the next geo-fiber as a scene change frame when the number of intra macroblocks in all P-frames in the geo-fi is less than a threshold in the fifth sub-step, and And if the number of intra macroblocks in all P-frames in the geo-fi is less than a threshold in the fifth sub-step, the seventh sub-step includes proceeding to the fifth step. Scene change detection method. The method of claim 1, wherein the fourth step, In the case where the geophysical structure is a non-closed geophysical sub-section, P-frames / I-frames included in a section where scene change possibility is detected and B-frames encoded by referring to the P-frames / I-frames as future images The first sub-steps that are tied together; A second sub-step of detecting the B-frame as a scene change frame when the P-frame constant is greater than a threshold and the B-frame constant is greater than a threshold in the sub-section bounded by the P-frame and the B-frames, A third sub-step of detecting the P-frame as a scene change frame if the P-frame constant is greater than a threshold and all of the B-frame constants are smaller than a threshold in the sub-section bounded by the P-frame and the B-frames; , A fourth sub-step of comparing the number of intra macroblocks of all P-frames with a threshold when all P-frame constants are smaller than a threshold in the interval, If the number of intra macroblocks of all P-frames is less than a threshold value in the fourth sub-step, a fifth sub-step of detecting a scene change frame in a sub-section concatenated with I-frames and B-frames, And a sixth sub-step proceeding to the fifth step if the number of intra macroblocks of all the P-frames is not less than a threshold in the fourth sub-step. The method of claim 1, wherein the special editing effect detection algorithm of the fifth step, A first sub-step of setting a special editable effect insertion section when the number of intra macroblocks of the P-frame is larger than a predetermined threshold value and the slope of the change amount of the number changes abruptly; A second sub step of obtaining a DC image for each frame of the insertable section of the special editing effect set in the first sub step; A third sub step of partially labeling the DC image; A fourth sub-step of comparing the DC images changed in the insertable section of the special editing effect with each other, A fifth sub-step in which the dissolve is inserted when the amount of change of the DC images in all the labeled regions is smaller than the set threshold in the fourth sub-step, And a sixth sub-step of determining that the wipe is inserted when the number of labeled regions in which the change amount is greater than a threshold is gradually increased in the fourth sub-step. In the method for detecting the beginning and end of the TV news article interposed between the anchored video (anchor frame), A first step of setting a candidate anchor frame if the face color is greater than or equal to a threshold value at the position where the anchor face appears according to the type of the anchor frame; Obtaining a DCT energy distribution of DC coefficients and AC coefficients for the DCT blocks constituting the candidate anchor frame; A third step of dividing the candidate anchor frame into a face portion and a background portion to obtain an average and a variance of DCT energy distribution of each block, respectively; A fourth step of measuring similarity between the candidate anchor frame and the anchor frame by comparing the average and the variance of the DCT energy distribution of the candidate anchor frame with the average and the variance of the DCT energy distribution of the anchor frame, respectively, Setting a candidate anchor frame as an anchor frame according to the similarity; And a sixth step of detecting between the anchor frames as an article region.