KR20000033070A - Apparatus for blocking broadcasting program watching using face recognition - Google Patents

Abstract

PURPOSE: An apparatus for blocking broadcasting program watching using face recognition is provided to block user from watching television selectively according to user's face. CONSTITUTION: An apparatus for blocking broadcasting program witching using face recognition includes a decoder, a recognizer(100), and a controller(30). The decoder encodes a bit stream from the broadcasting program and detects the level accompanied by the bit stream of watching. The recognizer(100) recognizes characteristics of members, and generated recognized result with identification of the member. The controller(30) generates an acknowledge signal and controls the decoder in an enable and a disable status according to the recognition result generated the face recognizer(100).

Description

Device for restricting viewing of broadcast programs using face recognition

The present invention relates to a method for restricting viewing of a broadcast program of digital television, and more particularly, to a method of selectively limiting viewing of a broadcast program of digital television using face recognition.

Along with the development of the media, various programs are being provided, and along with this, video programs harmful to minors, such as pornography, violence, and low-end broadcasting, are flooded. Various methods have been proposed to protect minors and youth from these harmful broadcast programs. One of them is a method of blocking a program being played or broadcast by using a password in a broadcasting medium, that is, a video cassette recorder and a television. However, this method of using passwords is not convenient for memorizing passwords as long as only one or two members of the family have the authority to block harmful programs. There is a problem that you can not use.

Another method for restricting viewing of a broadcast program is to find a language related to pornography and violence from a broadcast program provided by the broadcast system and to block a broadcast program including the language in advance. However, this method also can not accurately find harmful broadcast programs, the perfect program for this situation is not yet universal.

It is therefore an object of the present invention to provide a method for selectively limiting viewing through face recognition of a broadcast program viewer of digital television.

The apparatus for limiting broadcast program viewing of digital television using face recognition according to the present invention for achieving the above object, decodes the bit stream of the broadcast program and outputs it in a viewable form, and viewing the broadcast program accompanying the bit stream. Decoding means for detecting a rating; Member face recognition means for recognizing a feature of each member's face who wants to watch the broadcast program, and generating a confirmation result of confirming the identity of the member based on the recognized feature in response to a broadcast program viewing restriction signal. ; Generating the broadcast program parental control confirmation signal by comparing the parental level detected by the decoding unit with a preset parental level, and enabling or disabling the decoding unit in accordance with the verification result generated from the face recognition unit; It characterized in that it comprises a control means for controlling.

1 is a schematic block diagram of a broadcast program viewing restriction apparatus using face recognition constructed in accordance with a preferred embodiment of the present invention;

2 is a detailed block diagram of a face recognition unit illustrated in FIG. 1;

3 is a view for explaining a feature vector extracted by the feature vector extracting unit illustrated in FIG. 2;

<Explanation of symbols for the main parts of the drawings>

10: tuner / demodulator 20: MPEG decoder

30: TV control unit 100: face recognition unit

110: feature vector extractor 140: neural network

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

1, which is a schematic block diagram of a digital television for limiting viewing of a broadcast program using face recognition constructed in accordance with the present invention, the tuner / demodulator 10, the MPEG decoder 20, and the television controller 30 , A display 50, a face image acquisition unit 60, and a face recognition unit 100.

As shown in Fig. 1, a bit stream of a digital broadcast signal broadcast from a broadcasting station is applied to the tuning / demodulation section 10 of the digital television. The bit stream contains not only content information such as broadcast time, version, channel information, etc. of the broadcast program, but also a class signal indicating the class of the program. The program grade represented by this grade signal indicates an unrestricted age limit.

The tuner / demodulator 10 tunes a bit stream of a channel corresponding to the tuning control signal of the television control unit 30 among the bit streams of the digital broadcast signal received through the antenna, and then demodulates it and then decodes the MPEG decoder 20. The program grade is detected and provided to the television control unit 30.

The MPEG decoder 20 separates and decodes a broadcast signal provided from the tuner / demodulator 10 into an audio signal, a video signal, and various program information, and then decodes the digital video signal in a frame unit through a signal processor (not shown). Is converted into an NTSC or PAL signal and displayed on the display 50, and the digital audio signal decoded from the MPEG decoder 20 is converted into an analog signal and output to the speaker (not shown).

The television control unit 30 functions to control all the components constituting the digital television, and in connection with the face recognition unit 100 described below, a viewer who is currently watching television can watch a currently received broadcast program. By selectively controlling the MPEG decoder 30 to prevent the decoded broadcast program from being output to the display 50. More specifically, the television control unit 30 compares the program grade provided from the tuner / demodulation unit 10 with a preset parental level. As a result of the comparison, if the program grade is the same as the preset parental level, the viewer restriction check signal is output to the face recognition unit 100 to determine whether the viewer who wants to watch the currently received broadcast program is the viewer of age limit. .

In response to the viewing restriction confirmation signal transmitted from the control unit 30, the face recognition unit 100 determines whether the viewer who is currently operating or watching the television in front of the television is a viewer who is viewable or viewable through the face recognition. In addition, the determination result is notified to the control unit 30 so that the control unit 30 controls the currently received program from being output from the MPEG decoder 20 to the display 50.

2 illustrates a detailed block diagram of the face recognition unit 100. The face recognition unit 100 according to the present invention includes a feature vector extractor 110, a hidden Markov model parameter generator 120, a conditional probability calculator 130, and a neural network 150. The face recognition unit 100 according to the present invention should first recognize a face for each family member, for example, as a viewer group. Accordingly, the face recognition unit 100 acquires a face image of a member of the viewer group to be recognized using the face image acquisition unit 60 illustrated in FIG. 1, and performs a process of storing features of the acquired face image. do.

The face image acquisition unit 60 acquires, for example, a face image for each member of the viewer group using a CCD camera (not shown). In image acquisition, a median filter consisting of a low pass filter is used to remove dust or noise from the camera lens.

The face image per member acquired by the face image acquisition unit 60 is provided to the feature vector extractor 110 of the face recognition unit 100. In the following embodiment of the present invention, one family member is described, assuming that the number of face images acquired per person is 10. In this regard, the number of subjects to be recognized, the number of face images per subject and the number of feature vectors described below may be any number as long as it is sufficient to train the neural network 150.

The feature vector extractor 110 extracts an edge face image 200 as illustrated in FIG. 3 by a predetermined edge extraction technique and generates a feature vector. The edge facial image 20 shown in FIG. 3 includes two eyes 212 and 214, a nose 216 and a mouth 218, and feature vectors are generated based on these. The feature vector has five elements, [F1, F2, F3, F4, F5] = [a / f, b / f, c / f, d / f, e / f]. Where "a" is the diagonal length between the bottom midpoint of the left eye 212 and the bottom midpoint of the nose 216; "b" is the diagonal length between the bottom midpoint of the right eye 214 and the bottom midpoint of the nose 216; "c" is the vertical length between the bottom midpoint of nose 216 and the top midpoint of mouth 218; "d" is the vertical length between the bottom midpoint of the right eye 214 and the bottom midpoint of the nose 216; "e" is the horizontal length between the left midpoint of mouth 218 and the point from which the horizontal line meets the outline of face image 200; "f" is the horizontal length between the right side midpoint of left eye 212 and the left side midpoint of right eye 214. Five elements from "a" to "e" are defined as feature vectors (O _v = O ₁ , O ₂ , O ₃ , O ₄ , O ₅ ) by dividing and normalizing by "f".

As illustrated in Table 1, since the feature vectors O _v have 10 face images for every 10 members as described above, there are 5 feature vectors for each face image of each member. Therefore, since each member has a total of 50 feature vectors in 10 face images, there are a total of 500 feature vectors for all 10 members.

Target person Face image Feature vector Model parameters Member 1 Image 1 ... Image 10 O _v , ₁ = O ₁ , O ₂ , O ₃ , O ₄ , O ₅ ... O _v , ₂₀ = O ₁ , O ₂ , O ₃ , O ₄ , O ₅ λ ₁ ... ... ... ... Member 10 Image 1 ... Image 10 O _v , ₁ = O ₁ , O ₂ , O ₃ , O ₄ , O ₅ ... O _v , ₂₀ = O ₁ , O ₂ , O ₃ , O ₄ , O ₅ λ ₁₀

As illustrated in Table 1, 250 feature vectors O _v of 5 face images among 10 face images extracted by the feature vector extractor 20 are hidden Markov model parameter generators ( 120). The feature vector provided to the hidden Markov model parameter generator 120 is used as training data.

The hidden Markov model used in the present invention is a statistical model that can represent a signal as a double stochastic process, and has a sequence of features O ₁ , O ₂ ,. . . , O _Τ and the hidden state strings s ₁ , s ₂ ,. . . , s _Τ The hidden Markov model is a collection of states that are connected to each other by transition, and each transition involves two probabilities. One is the transition probability that the current transition is to be made, and the other is the probability that defines the conditional probability that each output symbol is observed from a finite number of observation objects when the transition is made.

This hidden Markov model has a state number of N, an observation symbol number of M, and a state set Q = q ₁ , q ₂ ,. . . . , q _N ｝ , Symbol set V = ｛v ₁ , v ₂ ,. . . . , v _M ｝ , The length of the observation column t = 1, 2,. . . , Γ It may be defined as a model parameter λ as shown in Equation 3 below.

λ = (A, B, π)

In Equation 3, the variable A is the state transition probability at which the t th observation symbol string is in state q _i , the t + 1 th observation symbol string is the state q _j to be selected, and the variable B is the t th observation symbol string is the state. The observed symbol probability to select the symbol v _k in the q _j state, and the variable π is an initial state probability to select the state q _j in the initial state, which are defined as Equations 4 to 6, respectively.

A = ｛a _ij ｝, a _ij = Pr (q _j at t + 1, q _i at t)

B = ｛b _j (k)｝, b _j (k) = Pr (v _k at t, q _j at t)

π = ｛π _i ｝, π _i = Pr (q _j at t = 1)

According to the present invention, in the hidden Markov model, five feature vectors O _v are used as the length of the observation sequence, the number of states is four, and a model parameter is generated by using a left-to-right model of two possible transition numbers. It was.

Therefore, the hidden Markov model parameter generation unit 30 is trained to receive five feature vectors O _v for each of five facial images per member and generate a model parameter λ that converges to the smallest value. As a result, a model parameter λ _i is generated that converges to one smallest value per person, and a total of 10 model parameters λ _i = λ ₁ , λ ₂ , λ ₃ ,. , λ ₁₀ ) is generated.

Then, the conditional probability generating unit 130 performs the converged 10 model parameters (λ _i = λ ₁ , λ ₂ , λ ₃ ,..., Λ ₁₀ ) and the remaining half of each member illustrated in Table 1. Conditional probability of the feature vector O _v for each member using the feature vector, Pr (O | λ) Find,. This conditional probability outputs the member having the largest probability value for each member as a recognition result, as shown in Table 2 below.

Target person Face image Model parameters Conditional probability Member 1 Image 1 ... Image 10 λ ₁ Pr (O _v , ₁ | λ ₁ ), Pr (O _v , ₁ | λ ₂ ),. . . , Pr (O _v , ₁ | λ ₁₀ ) ... Pr (O _v , ₁₀ | λ ₁ ), Pr (O _v , ₁₀ | λ ₂ ),. . . , Pr (O _v , ₁₀ | λ ₁₀ ) ... ... ... ... Member 10 Image 1 ... Image 10 λ ₁₀ Pr (O _v , ₁ | λ ₁ ), Pr (O _v , ₁ | λ ₂ ),. . . , Pr (O _v , ₁ | λ ₁₀ ) ... Pr (O _v , ₁₀ | λ ₁ ), Pr (O _v , ₁₀ | λ ₂ ),. . . , Pr (O _v , ₁₀ | λ ₁₀ )

As can be seen in Table 2 above, the conditional probabilities obtained for each member are calculated as 100 conditional probability values for each member. Of these, half the conditional probability value for each member is applied to the next neural network 150. The half conditional probability value applied to the neural network 150 is used as data for training the neural network 150, and the other half conditional probability value is stored therein for testing the learned neural network 150.

The neural network 150 is composed of an input layer, a hidden layer, and an output layer, as is widely known. In the present invention, the number of input nodes of the neural network 150 is 100 equal to the number of conditional probability values provided by the conditional probability generating unit 130, and the output node is composed of 10 equal to the number of members. The neural network 150 is trained based on the probability value provided by the conditional probability generating unit 130, so that the neural network 150 has a first value at the i th output node, i. 2 values, i.e., "0", are learned. The learning of neural network 40 is performed using conventional learning algorithms, for example, error back propagation algorithms.

Then, the conditional probabilities stored in the conditional probability generator 130 are sequentially input to the neural network 150 and the target person, that is, the viewers, are recognized based on the recognition result of the neural network 150.

Referring back to FIG. 1, when the control unit 30 of the digital television analyzes the parental level included in the broadcast program and provides the broadcast program parental control confirmation signal to the face recognition unit 100, the face image acquisition unit 60 is provided. When the face image of the person to be recognized who is currently watching TV is acquired, and the feature vector extracted from the feature vector of the face image of the person to be recognized by the feature vector extractor 110 is applied to the face recognition unit 100, the feature is extracted. The vector is input directly to the conditional probability generator 130 without passing through the hidden Markov model parameter generator 120 of the face recognition unit 100. The conditional probability generator 130 generates the conditional probability of the feature vector using the hidden Markov model. The generated conditional probabilities are then input to the learned neural network 150. The neural network 150 generates an output indicating which of the ten people the person corresponding to the feature vector is recognized.

As described above, by automatically and selectively executing viewing of a broadcast program viewer using face recognition, a member who has a specific authority to restrict viewing of a broadcast program does not have to worry about forgetting a password as in the prior art. Do not need to know the number.

Claims (3)

In the broadcast program viewing restriction device of digital television using face recognition, Decoding means for decoding and outputting a bit stream of the broadcast program in a viewable form, and detecting a broadcast program viewing grade accompanying the bit stream; Face recognition means for recognizing a feature of a face of each member who wants to watch the broadcast program, and generating a recognition result of confirming the identity of the member based on the recognized feature in response to a broadcast program viewing restriction confirmation signal; Comparing the parental level detected by the decoding means with a preset parental level to generate the broadcast program parental control confirmation signal, and enabling or disabling the decoding unit in accordance with a recognition result generated from the face recognition unit. And a control means for controlling the broadcast program viewing restriction apparatus using face recognition. The apparatus of claim 1, wherein the face recognition means comprises: A camera for acquiring a face image of a person to be recognized who wants to watch the broadcast program output by the decoding means; A feature vector extracting unit extracting a feature vector from a face of a member acquired from the camera; Using a face recognition, characterized in that it comprises a neural network for identifying the identity corresponding to the feature vector of the person to be recognized according to the results learned using the feature vector of each member to provide the recognition result to the control means Program viewing restriction device. The apparatus of claim 2, wherein the face recognition means comprises: A hidden Markov model parameter generator for generating a hidden Markov model parameter having the smallest convergence value by training using the feature vectors of each member; The conditional probability generator further generates a conditional probability of the feature vector by using the hidden Markov model parameter generated by the hidden Markov model generator and the feature vector of each member and provides the neural network to the neural network. Broadcast program viewing restriction device using face recognition, characterized in that.