TW201504955A - Method of detecting news anchorperson shot using face recognition - Google Patents

Abstract

Provided is a method of detecting news anchorperson shot using face recognition, which uses the recognition technique of anchorperson face to detect each anchorperson shot of the news program. The method provided by the invention includes two major steps: (1)anchorperson face learning: Within a section of the news program in the beginning stage the face model of an anchorperson is trained through an unsupervised learning, and used for determining whether the facial image is a criteria as for a news anchorperson's face; (2) anchorperson shot detection: after the step (1) is finished, according to the anchorperson face model and nchorperson face standard, whenever the anchorperson face appeared in the news program is determined, therefore, each of the anchorperson shots and its starting playback time in the news program is detected. The information can be used to separate a news program by unit of report.

Description

Method for detecting news anchor picture by using face recognition

The invention relates to a method for detecting an anchor picture in a news program, in particular to a news anchor picture detection technology combining image processing and visual content analysis. The method proposed by the invention is a key feature of the anchor image as the main broadcast screen, so the face recognition technology is needed to identify the face appearing in the news. In addition, the present invention, in conjunction with unsupervised learning techniques, allows the operation of the proposed method to be fully automated.

The anchor picture is a special scene in the news program. Generally speaking, a news program is mainly composed of dozens of news reports, and each news report is composed of a news anchor picture and an immediately interviewed reporter.

The basic concept of detecting the anchor picture is to detect the important features and characteristics of the news anchor picture. The common methods can be roughly divided into two types, one is scene change detection, which is to find the turning point of the anchor screen and the interview screen from the news. The principle is that the anchor picture has a special picture structure, including a fixed anchor and a background image with little change, which is more stable than the interview picture taken in various occasions. So when news changes from a changing scene to a stable scene, it is very likely that the anchor picture will appear. Therefore, the turning point from the changing scene to the stable scene and the transition from the stable scene to the changing scene The point is the main broadcast picture; the other method is scene clustering, which analyzes the complete news program, clusters similar pictures, and then filters out the anchor picture. The principle of this method is that since the anchor picture has a fixed picture structure, the anchor pictures belonging to different reports still have a relatively high degree of similarity, so they can be gathered together by the grouping process. Then, through the characteristics that the anchor picture has repeated appearance according to a certain frequency, it is possible to find out the group of pictures that are actually composed of the anchor pictures to achieve the effect of detecting the anchor picture.

However, in recent years, with the evolution of technology, the above two methods have emerging new shortcomings. First, the current news screen is more fancy and changeable than in the past. The traditional method is based on the identification that the anchors are kept in the same position in the anchor picture, and the background part behind them is fixed or only slightly changed. However, observing the current news program, the location of the anchor in the same news program will change, and the animation of the interview screen and the animation effect of the report have replaced the scene of the past system as part of the background of the anchor screen. Therefore, the method of finding the anchor picture by detecting the degree of change of the background picture will greatly reduce the accuracy. For example, in the case of the former Republic of China patent M386559 and the US patent 7720281B2, the human skin color detection is used to exclude the block of the news picture that may be the main broadcast face, and then the scene change detection method is used to find out the change of the remaining background picture. News anchor screen. Moreover, the skin color of the method must be defined in advance, but the skin color may vary with different news signal sources, which may affect the stability of the method performance. In addition, the traditional scene clustering method is not suitable for instant news streaming because it must analyze the picture distribution of a complete news program. However, as smart TVs gradually become a new generation of TV products, unlike the past, smart TVs have the function of connecting networks and interacting with humans. This makes the audience no longer just the party receiving the message, and thus can have more instant interaction with the TV. If you want to provide an audience, When the news content interaction function, it is necessary to analyze the news with a technology that is instantaneous and does not require human intervention. The technology that needs to wait until the entire news has been played to begin analysis is not up to the requirements.

It can be seen that there are still many shortcomings in the above-mentioned methods of use, which is not a good design, but needs to be improved.

In view of the shortcomings derived from the above-mentioned conventional methods, the inventor of the present invention has improved and innovated, and after years of painstaking research, he finally succeeded in researching and developing the method for detecting the news anchor picture using face recognition.

The object of the present invention is to propose a method for detecting an anchor picture for a news program. The system can analyze the instant news stream, and synchronously detect each of the anchor pictures and their initial playing time as the news is played.

A second object of the present invention is to provide a method for detecting a news anchor picture, which can detect the anchor picture without analyzing the entire frame, so that the stable characteristics of the anchor picture are less and less obvious. phenomenon.

A further object of the present invention is to provide an unsupervised learning method for a news anchor face model, which can utilize a segment of a news program to automatically and instantly train the anchor of the news without human intervention. Face model.

The method for detecting the news anchor picture by using face recognition to achieve the above object is to use the anchor face recognition technology to detect each anchor picture in the news. Because the anchor's face is a fairly stable key feature in the main broadcast and is not interfered by the background image. The appearance of the anchor picture can be detected very quickly by the face of the anchor. In order to determine whether a news frame contains an anchor face, it is necessary to first find all the face images in the frame through the face detection technology, and further identify whether the face image is the face of the main broadcast. In the process of the anchor face recognition, the face model of the news anchor is needed. If the difference between a face image and the anchor face model is less than the error range, the face can be identified as The anchor face. In order to obtain the anchor face model, the method proposed by the present invention is to start a segment by using a news program and automatically train it through the unsupervised learning method. After that, as the news is played, whenever there are several consecutive frames containing the anchor face, it means that there may be the appearance of the anchor picture.

The method proposed by the invention comprises two major steps: (1) anchor face learning: first capturing a news program to start all face images in a segment, and then analyzing the image to filter out the most likely anchor face. Image and use this to train the anchor model of this news. After obtaining the anchor face model, all the face images in the next segment of the news are captured, and then the difference between the face images and the anchor face model is analyzed to further evaluate whether a face image is (2) anchor screen detection: After completing step (1), use the anchor face model and the anchor face standard to identify the news in the broadcast, when there will be a frame containing the anchor face. . Whenever a frame containing an anchor face appears frequently for a period of time, it means that there is an anchor picture during this time, and the initial play time of the anchor picture is estimated.

A method for detecting a news anchor picture by using face recognition, the steps comprising: a. anchor face learning, using a segment starting from a news program, and instantly training the anchor face of the news program by using an unsupervised learning method. Model and anchor face standard; and b. anchor screen detection, using the anchor face model and the anchor face standard obtained from step a, identify the anchors that appear in the news in order to instantly detect the anchor screen Starting with time.

The main body face learning includes the following steps: c. anchor face model training, which uses the face image in a segment of the news program to automatically filter out the face image of the anchor to train the news program. The anchor face model; and d. the anchor face standard evaluation, using the anchor face model obtained in step c and the face image in another news program segment to evaluate a face image as the host face Standard.

The method for training the anchor face model includes: capturing a frame in the news image signal through a news image receiving component; and using a face detection component to extract a frame by using a face detection technology; All of the face images; and through an anchor face model training component, the host face model is trained using a plurality of face images obtained from the face detection component.

The main body face standard evaluation includes the steps of: capturing a video frame in a news image signal through a news image receiving component; and using a face detection component to extract a frame by using a face detection technology; All face images; and an anchor face standard evaluation component, using a plurality of face images and anchor face models obtained from the face detection component to evaluate the anchor face criteria.

The method for detecting the anchor image includes: capturing a video frame in the news image signal through a news image receiving component; and using a face detection component to extract a video in the image frame by using a face detection component All face images; Through an anchor frame recognition component, the anchor face recognition technology is used to determine whether a frame is a frame of a face image containing the news program anchor, that is, an anchor frame; and the anchor frame is used to detect the anchor through an anchor picture detection component. The initial playback time of the screen.

The main frame identification component includes the following steps: e. capturing image feature values of all face images in a frame; f. calculating each face image and the anchor face based on the image feature values After the difference value of the model, the standard score of the dissimilar value is calculated by using the anchor face standard; and g. the standard score is used to identify whether the face image is the main broadcast face, and accordingly, the frame containing the anchor face is set as the main broadcast. Frame.

The main picture detection component includes a processing step including: h. storing the received frame type of a frame in a variable, and storing the frame type of the past N _D frames including the frame; i. Calculate the number of anchor frames in the variable R; j. Check if the condition of R>N _D /2 is true; k. When the condition described by j is established, it means that an anchor picture is detected, so the current anchor picture is detected. The measured state is set to the anchor state, otherwise it is set to the non-anchor state; and l. When the condition described by j is established and the anchor screen detection state is converted from the non-anchor state to the current anchor state, the earliest one of the output variables is the anchor. The playback time of the frame is the starting time of the playback of this anchor screen.

The status of the anchor picture detection refers to two states in the process of detecting the anchor picture, including: The news is currently in the anchor state of the anchor screen; and the news is currently not in the non-hosting state of the anchor screen.

The anchor picture detection state is in a non-host state.

The anchor face model training component includes the following steps: m. capturing image feature values of the plurality of face images; n. using the grouping method to image all face images according to similarity of image feature values Divided into three groups; o. For each face image group, calculate the group degree of group value, which is defined as the number of face images within the group divided by the difference between the image feature value in the group and the average image feature value. Standard deviation; and p. Select a face image in the face image group with the largest clustering degree to train the anchor face model.

The main face standard evaluation component includes the following steps: q. capturing image feature values of the plurality of face images; r. calculating each face image and the anchor person based on the image feature values Different values of the face model; s. use the grouping method to divide all the different values into two groups and calculate the average of the different values; and t. select a group with the average of the lower dissimilar values, and From this, the anchor face standard is evaluated, which includes the average value of the different values and the standard deviation of the different values.

The method for detecting the anchor image includes: capturing a video frame in the news image signal through a news image receiving component; and using a face detection component to extract a video in the image frame by using a face detection component All face images; through an anchor frame recognition component, using the anchor face recognition technology to determine whether a frame is included The image of the face image of the news program anchor, that is, the anchor frame; and the anchor picture detection component, uses the anchor frame to detect the anchor picture and its initial play time.

The method for detecting a news anchor picture by using face recognition disclosed in the present invention has the following advantages when compared with other conventional technologies:

1. The anchor picture detection method of the present invention can synchronously detect the anchor picture along with the news broadcast, so that the instant interaction function with the news can be provided, such as the news fast reverse function by a single report.

2. The method for detecting the anchor picture of the present invention only uses the anchor face as a key feature for finding the anchor picture without analyzing the news background picture, so that it can cope with an increasingly complicated news picture.

3. The method for detecting the anchor picture of the present invention is to confirm the appearance of the news anchor by the face recognition technology, so that the performance of the anchor is more stable than the low-order image features such as color.

4. The anchor face learning method of the present invention can automatically train the anchor face model of a news program in an unsupervised learning method, so that the method is not suitable for the application of the method to an instant operation system.

5. The anchor face learning method of the present invention can instantly train the anchor face model of a news program, so that the anchor face model can be surely consistent with the current anchor face.

The detailed description of the preferred embodiments of the present invention is intended to be limited to the scope of the invention, and is not intended to limit the scope of the invention. The patent scope of this case.

In summary, this case is not only innovative in terms of space type, but also can enhance the above-mentioned multiple functions compared with the conventional items, and should fully comply with the statutory invention patents of novelty and progress. If you apply in accordance with the law, you are requested to approve the application for the invention patent in order to invent the invention.

11‧‧‧Hosting face learning

12‧‧‧Host screen detection

21‧‧‧Hosting face model training

211‧‧‧News image receiving component

212‧‧‧Face Detection Component

213‧‧‧Host Face Model Training Component

22‧‧‧Audit face standard assessment

221‧‧‧News image receiving component

222‧‧‧Face Detection Component

223‧‧‧Host Face Standard Evaluation Component

51‧‧‧News image receiving component

52‧‧‧Face Detection Component

53‧‧‧Annual frame recognition component

54‧‧‧ anchor screen detection component

S1~S24‧‧‧Step process

FIG. 1 is a flow chart of a method for detecting an anchor picture by using face recognition according to the present invention.

Figure 2 is a flow chart of the main face learning in the present invention.

Figure 3 is a flow chart of the training in the anchor model training component of the present invention.

Figure 4 is a flow chart of the evaluation in the anchor face standard evaluation component of the present invention.

Figure 5 is a flow chart of the anchor picture detection in the present invention.

Figure 6 is a flow chart of identification in the anchor frame recognition component of the present invention.

Figure 7 is a flow chart of detection in the anchor picture detecting component of the present invention.

Figure 8 is a diagram showing the state change of the anchor picture detection in the present invention.

The technical features, contents, and advantages of the present invention, as well as the advantages thereof, can be understood by the reviewing committee, and the present invention will be described in detail with reference to the accompanying drawings. The subject matter is only for the purpose of illustration and description. It is not intended to be a true proportion and precise configuration after the implementation of the present invention. Therefore, the scope and configuration relationship of the attached drawings should not be interpreted or limited. First described.

Please refer to FIG. 1 , which is a flowchart of a method for detecting an anchor screen by using face recognition according to the present invention. The method includes two major steps: anchor face learning 11 and anchor screen detection 12 .

S1: First, the news program starts to play;

S2: The anchor face learning 11 will use a segment of the news program to automatically train the anchor face model and the anchor face standard of the news program in an unsupervised learning manner;

S3: After completing the above S2 step, the anchor screen detection 12 utilizes the anchor face model and the anchor face standard obtained from the anchor face learning 11 to instantly detect each of the appearing anchor screens with the news broadcast. Start time

S4: The news program ends playing and the detection action stops.

Please refer to FIG. 2 , which is a flowchart of the method for the anchor face learning of the method for detecting the anchor picture by using the face recognition method of the present invention, which comprises two steps:

S5: First, the news image receiving component 211 in the anchor face model training 21 sequentially captures the playing news frame, and sends the frame to the face detecting component 212 to extract all face images in the frame. . The extracted face image is transmitted to the anchor face model training component 213. When the anchor face model training component 213 to obtain N _T1 faces in images, news video receiver assembly 211 to stop extracting news Frame, and anchor face model training component 213 from N _T1 instantly to human faces in the image The unsupervised learning method trains the anchor face model. Where N _{T1 is the} number of face images used to train the anchor face model, and its definition is as shown in equation (1): N _{T 1} = s ₁ * t ₁ (1)

Where s1 is the number of frames captured by the news image receiving component 211 in each second of the news, and t1 is the number of seconds of the face image required. After completing the anchor face model training 21, an anchor face model is obtained;

S6: The news image receiving component 221 in the anchor face standard evaluation 22 sequentially captures the news frames played after, and sends the frames to the face detecting component 222 to extract all face images in the frame. The extracted face image is transmitted to the anchor face standard evaluation component 223. When the anchor face evaluation component 223 in the standard face image to obtain N _T2 sheets, news video receiver assembly 221 to stop capturing Frame press, while the face evaluation component 223 would use standard S5 of the resulting face model with this anchor N _{The T2} face image automatically evaluates the anchor face standard. Where N _{T2 is the} number of images used to evaluate the standard of the anchor face, and its definition is as shown in equation (2): N _{T 2} = s ₂ * t ₂ (2)

Where s ₂ is the number of frames captured by the news image receiving component 221 in the news image of each second, and t ₂ is the number of seconds of the face image required. After completing the anchor face criteria evaluation 22, a set of anchor face criteria will be obtained.

Please refer to FIG. 3, which is a training flowchart in the anchor face model training component of the method for detecting the anchor picture by using the face recognition method of the present invention.

S7: after obtaining the N _T1 face images, first capturing the image feature values of the face images: a local binary pattern (LBP);

S8: Based on the similarity of the feature values of the face image, all the face images are divided into three groups by k-means clustering. The similarity of the image feature values is defined as the Euclidean distance between the feature values, and the smaller the distance, the higher the similarity;

S9: Next, for each face image group, the cluster degree value C(m) of the group is calculated, and the definition is as shown in the formula (3):

Where m=1, 2, 3, N _m is the number of face images in the m-th group face image, and std _m is the degree of dispersion of the image feature values in the m-th group face image, and the definition is as follows Equation (4):

Where d _mi is the Euclidean distance of the feature image value of the i-th face image in the m-th group face image and the average image feature value of the group, and D _m is all in the m-th group face image The average of d _mi . Wherein, the value of the jth dimension of the group average image feature value X _{m,j is} defined as shown in the formula (5):

Where x _mi,j represents the value of the j-th dimension of the image feature value of the i-th face image in the m- _th group face image. Because in a news program, the anchor image will appear frequently and the similarity between the images is high, the larger the C (m) of a face image group, the more likely it is to be an anchor. a group of face images composed of faces;

S10: Finally, a group of face images with the largest degree of clustering is selected from the three groups of face images, and the face model of the news program is trained using the face image. The image feature value of the anchor face model in the present embodiment also uses LBP.

Please refer to FIG. 4, which is an evaluation flowchart of the anchor face standard evaluation component of the method for detecting the anchor picture by using the face recognition method of the present invention.

S11: After obtaining the N _T2 face images, the image feature values LBP of the face images are captured.

S12: Calculate a difference value between each face image and the anchor face model by using image feature values, which is defined as a Euclidean distance between image feature values. When the difference value is greater, it represents this person. The lower the similarity between the face image and the anchor face model;

S13: Then, using the K-average grouping algorithm, all the dissimilar values are divided into two groups, and the average values of the disparity values of the two groups are respectively calculated. Because the difference between the anchor face and the anchor face model is lower, and the difference between the non-host face and the anchor face model is higher, the distribution of the two different values will be significantly different. ;

S14: Finally, a group with a lower average value is selected, that is, the difference between the anchor face and the anchor face model can be known, so the average value of the different values of the group is crit ₁ and the standard deviation crit ₂ Be the main broadcast face standard. Its definition is as shown in equations (6) and (7):

Wherein, N _T3 and e _i respectively represent the number of distinct values and the i-th different value in a group of average values having lower dissimilar values.

Please refer to FIG. 5 , which is a flowchart of the anchor picture detection method for detecting the anchor picture by using the face recognition method, and illustrates how to detect the anchor picture every time a news frame is received.

S15: The news image receiving component 51 is responsible for receiving the news signal in play and extracting the news frame, and then sending it to the face detecting component 52 in sequence;

S16: the face detection component 52 will extract all the face images in a frame and pass to the anchor frame recognition component 53;

S17: The anchor frame recognition component 53 uses the captured face image to determine whether the type of the frame is an anchor frame or a non-host frame, and transmits the result to the anchor picture detection component 54;

S18: Finally, the anchor screen detection component 54 will incorporate the frame type of the frame into a reference for detecting whether the current news is on the anchor screen. Once a new anchor picture is detected, the start time of this anchor picture is immediately output. However, if a detected anchor picture is detected, the initial playback time of the anchor picture is no longer output.

Please refer to FIG. 6 , which is a flowchart of identification in the anchor frame recognition component of the method for detecting the anchor image by using the face recognition method, and illustrates how to identify a frame as a main frame or a non-host frame. Among them, an anchor frame refers to a frame containing an anchor face image, and a non-host frame refers to a frame that does not contain a face image or a face image without an anchor face. In order to identify whether a face image is a host face image, the steps are as follows:

S19: First, the image feature value LBP of the face image is captured.

S20: Next, the image feature value is used to calculate the difference ef between the model and the anchor face model, which is defined as the Euclidean distance between the image feature values of the two. Then calculate the standard score (z-score) of this disparity value, which is defined as shown in equation (8):

S21: When z is less than a threshold value (the threshold) θ, then the face image is the main broadcast face, where θ is between 2 and 4. Finally, the type of the frame is determined according to the recognition result of the face image.

Please refer to FIG. 7 , which is a detection flowchart in the anchor image detecting component of the method for detecting the anchor image by using the face recognition method, which is used to detect whether an anchor screen appears when a frame is received. the process of.

S22: First, an anchor frame recognition result received from the anchor frame recognition component 53 is stored in a variable. This variable records the type of past N _D frames containing this frame. The size of N _D is defined as follows: N _D = s ₃ * t ₃ (9)

Where s ₃ is the number of frames captured by the news image receiving component 51 in each second of the news image, and t ₃ is the minimum number of seconds of the main broadcast picture, and the value is between 5 and 10.

S23: Calculate the number of anchor frames in this variable R

S24: Finally, according to the value R, the change of the detection state of the anchor picture is controlled, and when a new anchor picture is detected, the initial play time is output. Among them, the change of the status of the anchor screen is shown in Figure 8. The anchor screen detection process includes two states: the anchor state and the non-host state. Among them, the anchor status means that the current news is in the anchor picture, and the non-host status means that the current news is not in the anchor picture. Instead of the anchor state, the initial state of the main screen detection process. The change in the two states will change depending on the relationship between R and N _D . The start time of an anchor picture is output only when more than half of the frames in the variable are the main frame, ie R>N _D /2, and the status is changed from the non-online status detection status to the anchor status. The starting time of the anchor screen is the playing time of the earliest anchor frame in the current variable.

The method for detecting a news anchor picture by using face recognition disclosed in the present invention has the following advantages when compared with other conventional technologies:

1. The anchor picture detection method of the present invention can synchronously detect the anchor picture along with the news broadcast, so that the instant interaction function with the news can be provided, such as the news fast reverse function by a single report.

2. The method for detecting the anchor picture of the present invention only uses the anchor face as a key feature for finding the anchor picture. It is not necessary to analyze the news background image, so it can respond to increasingly complex news images.

3. The method for detecting the anchor picture of the present invention is to confirm the appearance of the news anchor by the face recognition technology, so that the performance of the anchor is more stable than the low-order image features such as color.

4. The anchor face learning method of the present invention can automatically train the anchor face model of a news program in an unsupervised learning method, so that the method is not suitable for the application of the method to an instant operation system.

5. The anchor face learning method of the present invention can instantly train the anchor face model of a news program, so that the anchor face model can be surely consistent with the current anchor face.

The detailed description of the present invention is intended to be illustrative of a preferred embodiment of the invention, and is not intended to limit the scope of the invention. The patent scope of this case.

To sum up, this case is not only innovative in terms of technical thinking, but also has many of the above-mentioned functions that are not in the traditional methods of the past. It has fully complied with the statutory invention patent requirements of novelty and progressiveness, and applied for it according to law. Approved this invention patent application, in order to invent invention, to the sense of virtue.

11‧‧‧Hosting face learning

12‧‧‧Host screen detection

S1~S4‧‧‧Step procedure

Claims (12)

A method for detecting a news anchor picture by using face recognition, the steps comprising: a. anchor face learning, using a segment starting from a news program, and instantly training the anchor face of the news program by using an unsupervised learning method. Model and anchor face standard; and b. anchor screen detection, using the anchor face model and the anchor face standard obtained from step a, identify the anchors that appear in the news in order to instantly detect the anchor screen The starting time of its start. The method for detecting a news anchor picture by using face recognition according to the first aspect of the patent application, wherein the method for the subject face learning comprises: c. anchor face model training, using a segment of the news program The face image automatically filters out the face image of the anchor to train the anchor face model of the news program; and d. The anchor face standard evaluation uses the anchor face model obtained in step c and another news The face image in the program segment evaluates the standard of a face image as the main face. The method for detecting a news anchor picture by using face recognition according to the second aspect of the patent application, wherein the method for training the anchor face model comprises: capturing, by a news image receiving component, the news image signal in sequence Video frame; through a face detection component, using face detection technology to extract all face images in a frame; and using a host face model training component, using a plurality of faces obtained from the face detection component The image trains the anchor face model. The method for detecting a news anchor picture by using face recognition according to item 3 of the patent application scope, wherein the method for assessing the subject face criteria includes: Through a news image receiving component, the video images in the news image signal are sequentially captured; through the face detection component, the face detection technology is used to extract all the face images in one frame; and through a host face standard The evaluation component uses a plurality of face images and an anchor face model obtained from the face detection component to evaluate the anchor face standard. The method for detecting a news anchor screen by using face recognition according to the fourth aspect of the patent application, wherein the step of detecting the anchor screen comprises: capturing, by a news image receiving component, the video image signal in sequence Video frame; through a face detection component, using face detection technology to extract all face images in a frame; through an anchor frame recognition component, using the anchor face recognition technology to determine whether a frame contains the news program anchor The image of the face image, that is, the anchor frame; and the anchor picture detection component, uses the anchor frame to detect the anchor picture and its initial playback time. The method for detecting a news anchor picture by using face recognition according to claim 5, wherein the main frame identification component comprises the following steps: e. capturing image features of all face images in a frame Value; f. based on the image feature value, after calculating the difference value of each face image and the anchor face model, using the anchor face standard to calculate the standard score of the different value; and g. identifying whether there is someone by the standard score The face image is the main broadcast face, and the frame containing the anchor face is set as the main broadcast frame accordingly. The method for detecting a news anchor picture by using face recognition according to the sixth aspect of the patent application, wherein the main picture detection component includes the following steps: h. storing the received frame type of a frame in a a variable containing the frame type of the past N _D frames including the frame; i. calculating the number of anchor frames in the variable R; j. checking whether the R>N _D /2 condition is true; k. when step j When the condition is met, it means that an anchor picture is detected, so the current anchor picture detection status is set to the anchor status, otherwise it is set to the non-host state; and l. The condition described in step j is established and the anchor picture is set. When the detection state is converted from the non-anchor state to the current anchor state, the play time of the earliest one of the output frames in the output variable is used as the starting play time of the anchor screen. The method for detecting a news anchor picture by using face recognition according to claim 7 of the patent application scope, wherein the anchor picture detection status refers to two states in the process of detecting an anchor picture, including: the news is currently in The anchor status of the anchor screen; and the news is currently not in the non-anchor state of the anchor screen. The method for detecting a news anchor picture by using face recognition according to item 8 of the patent application scope, wherein the anchor picture detection state is initially in a non-anchor state. The method for detecting a news anchor picture by using face recognition according to the fourth aspect of the patent application, wherein the anchor face model training component comprises the following steps: m. capturing respective images of the plurality of face images The eigenvalues; n. use the grouping method to divide all face images into three groups according to the similarity of image feature values; o. For each face image group, calculate the group degree of group value, which is defined as the group of people Face map The number of images divided by the standard deviation of the difference between the image feature value and the average image feature value; and p. Select a face image in the face image group with the largest clustering value to train the anchor face model . The method for detecting a news anchor picture by using face recognition according to claim 10, wherein the anchor face standard evaluation component includes the following steps: q. capturing respective images of the plurality of face images Eigenvalue; r. Calculate the difference value between each face image and the anchor face model based on the image feature value; s. Use the grouping method to divide all the disparate values into two groups and calculate the respective dissimilar values The average of; and t. select a group of averages with lower disparity values and evaluate the subjective face criteria from which the mean value of the disparity values and the standard deviation of the disparity values are included. The method for detecting a news anchor screen by using face recognition according to the first aspect of the patent application, wherein the step of detecting the anchor screen comprises: capturing, by a news image receiving component, the video image signal in sequence Video frame; through a face detection component, using face detection technology to extract all face images in a frame; through an anchor frame recognition component, using the anchor face recognition technology to determine whether a frame contains the news program anchor The image of the face image, that is, the anchor frame; and the anchor picture detection component, uses the anchor frame to detect the anchor picture and its initial playback time.