KR20150030445A - System for detecting child pornography - Google Patents

Abstract

Provided is a system for detecting child pornography, which is capable of dividing and detecting effectively general pornography and child pornography using a soft biometric recognition technology. The system includes a video providing device configured to store at least one video content and provide the same to the outside; and a child pornography determination device configured to extract the video content stored in the video providing device, to analyze the incontinence of the extracted video content to determine incontinence, and to analyze soft biometric identification information on each of people from objects included in the video content by using a soft biometric identification method to determine whether the video content corresponds to child pornography content.

Description

System for detecting child pornography [

The present invention relates to a pornographic object detection system, and more particularly, to a child pornographic object detection system capable of classifying child pornographic objects by using a soft biometric technique in multimedia data.

Child pornography has a great impact on sexual violence crimes and has recently become a serious social problem in that it is a pornographic object intended for children who should be protected as a socially weak person. Therefore, although punishment for distributing, possessing, and showing child pornography has been strengthened, there is no technology developed to classify and search adult pornography and child pornography in PC, USB memory, CD / DVD, streaming service and the like.

Techniques for detecting and detecting common pornography have been developed, for example, by generating and comparing hash values of multimedia data, and analyzing the contents of multimedia data. The method of using the hash value of the multimedia data has a disadvantage in that it can not be detected even if the obscene content included in the hash list includes codec, resolution change, image editing, and the like. Therefore, it is difficult to use it for the purpose of retrieving whether or not child pornography exists in the storage device.

In the method of detecting pornographic material by analyzing the contents of multimedia data, it is difficult to distinguish the pornographic material because the pornographic material is similar in characteristics to the pornographic material. Therefore, it is difficult to use the technique of classifying general pornography as limited use of child pornography detection.

Korean Patent No. 1013484 discloses a pornographic material detection system for detecting whether or not a pornographic object is an image through an object extracted using an image segmentation algorithm. However, this pornographic object detection system only detects pornographic material, .

SUMMARY OF THE INVENTION It is an object of the present invention to provide a child pornography detection system that can effectively detect general pornography and child pornography effectively using soft biometric technology.

According to an aspect of the present invention, there is provided a child pornography detection system comprising: an image providing device capable of storing at least one image content and providing the image content to the outside; And extracting at least one image content stored in the image providing device and analyzing the selectivity of the extracted image content to determine a degree of selectivity, and determining a degree of selectivity for each of the objects included in the image content using a soft biometrics technique A child pornography determining device for analyzing soft biometric information to determine whether the image content is child pornographic content; .

Therefore, the child pornography detection system of the present invention can improve the accuracy of child pornography detection by using information such as sex, age, and the like of characters created through image selection and soft biometry, The sex and age of the child can be used in combination with the kidney information and the voice information to enhance the accuracy of the sex and age estimation, thereby improving the child pornography classification performance.

In addition, it is possible to detect child pornography in streaming data by providing a method of detecting child pornography in a sequential manner, and to improve the retrieval speed when scanning a large number of files in a mass storage.

1 shows a child pornography detection system according to an embodiment of the present invention.
2 shows an embodiment of the selectivity analyzing unit according to the present invention.
3 shows another embodiment of the selectivity analyzing unit according to the present invention.
4 shows an embodiment of a soft biometric analysis unit according to the present invention.
FIG. 5 illustrates an example of application of soft biometric information analyzed by the soft biometric analysis unit of FIG.
Fig. 6 shows an embodiment of a child pornography classification unit according to the present invention.
7 shows an example in which a feature combining module generates a feature vector.
Fig. 8 shows another example in which the feature combination module generates the feature vector.
Fig. 9 shows an embodiment of a child pornography classification module according to the present invention.
FIG. 10 shows a method for detecting pornography of a child according to an embodiment of the present invention.
11 shows a method for detecting child sexual pornography according to another embodiment of the present invention.

In order to fully understand the present invention, operational advantages of the present invention, and objects achieved by the practice of the present invention, reference should be made to the accompanying drawings and the accompanying drawings which illustrate preferred embodiments of the present invention.

Hereinafter, the present invention will be described in detail with reference to the preferred embodiments of the present invention with reference to the accompanying drawings. However, the present invention can be implemented in various different forms, and is not limited to the embodiments described. In order to clearly describe the present invention, parts that are not related to the description are omitted, and the same reference numerals in the drawings denote the same members.

Throughout the specification, when an element is referred to as "including" an element, it does not exclude other elements unless specifically stated to the contrary. The terms "part", "unit", "module", "block", and the like described in the specification mean units for processing at least one function or operation, And a combination of software.

1 shows a child pornography detection system according to an embodiment of the present invention.

Referring to FIG. 1, the child pornography detection system of the present invention includes an image providing apparatus 100 and a child pornography determining apparatus 200.

The image providing apparatus 100 is configured to store at least one image content and provide the image content to the outside. The image providing apparatus 100 includes a local data storage 110 that can store image contents included in a smart phone, a smart pad, and a PC, a network storage 120 that can provide image contents through a network at a remote site, And a streaming service device 130 capable of providing real-time image contents using a streaming technique.

The child pornography determining apparatus 200 includes an image extracting unit 210, a sensibility analyzing unit 220, a soft biometric analyzing unit 230 and a child pornography sorting unit 240.

The image extracting unit 210 extracts at least one image content stored in the image providing apparatus 100 to acquire the input image content. The image extracting unit 210 detects at least one image content stored in an image storage medium such as the local data storage 110, the network storage 120 and the streaming service apparatus 30 of the image providing apparatus 100, At least one video content. The image extracting unit 210 senses at least one image content included in the image providing apparatus 100 using an extension of a stored file or collects and analyzes data output from the image providing apparatus 100 to the outside The image content can be detected. Then, the image extracting unit 210 transmits the received image content to the sensibility analyzing unit 220 and the soft biometric analyzing unit 230 to analyze the image.

At this time, the image extracting unit 210 may transmit all information of the voice and image included in the image content to the sensibility analyzing unit 220 and the soft biometrics analyzing unit 230, but in some cases, only a part of the image content And may be selectively transmitted by sampling. In the case of a general HD-class moving image, it is composed of about 24 to 30 frames per second. When the moving image is about one hour, it is about 80,000 frames. The sensibility analysis unit 220 and the soft biometrics analysis unit 230 . In addition, since there are many information overlapping in a continuous image such as a motionless image or a motionless image, sampling is required to efficiently analyze the image. Accordingly, the image extracting unit 210 may extract a scene by a sampling method based on a threshold value such as time unit sampling for extracting an analysis interval at predetermined time intervals, frame-by-sample sampling for extracting an analysis interval for each predetermined number of frames, Or a method of extracting an analysis section by detecting a change of a shot or the like and may transmit the sampled data to the sensibility analysis unit 220 and the soft biometrics analysis unit 230. For example, the image extracting unit 210 analyzes the image content, extracts a section including a person among the objects displayed in the image content as an analysis section, and transmits the extracted section to the sensibility analysis unit 220 and the soft biometrics analysis unit 230 Lt; / RTI > That is, since the present invention is a method for detecting child pornography, there is no need to analyze image content that does not include a person. Therefore, the image extracting unit 210 may perform the filtering operation in advance so that the selectivity analyzing unit 220 and the soft biometric analyzing unit 230 do not unnecessarily analyze the image contents.

The selectivity analyzing unit 220 receives the image content from the image extracting unit 210 and transmits information such as the characteristics (color, shape, texture), motion, and sound of the received image content to a frame unit or a predetermined analysis unit Analysis is made to determine the selectivity. That is, it is judged whether or not obscene material is present. The susceptibility analyzing unit 220 can be applied to both hash-based pornography detection method and content analysis-based pornography detection method used for existing pornographic material detection. As a result, the selection line analyzer 220 determines whether the image content is pornographic content, without distinguishing the pornography object from the general pornography object using the existing pornographic material detection method.

The soft biometrics analysis unit 230 receives the image content from the image extraction unit 210, determines whether a person is included in the object included in the received image content, and uses the soft biometrics technique Respectively.

Unlike traditional biometric technology, which is used to accurately identify an individual's identity, soft biometric technology analyzes characteristics naturally recognized by people (sex, age, height, skin color, gait, etc.) It is a technique used to acquire information. Traditional biometric technologies required user cooperation in a constrained environment to accurately identify an individual. For example, the conventional biometric technology corresponds to a technique of identifying a specific person among registered people by using fingerprint information, iris information, finger vein information, and the like in various security devices. In order to discriminate a specific person, information about each individual to be discriminated should be collected in advance. Therefore, it is necessary to provide users with fingerprint and iris information in a restrictive environment. . On the other hand, soft biometric technology is a technology for acquiring only the approximate characteristics of individual such as sex, age, height, etc., and does not require user cooperation. Therefore, it can be easily used for video surveillance and analysis of stored images.

Accordingly, the soft biometric analysis unit 230 can determine whether the person included in the image is an adult or a child using soft biometric technology.

The child pornography classifying unit 240 performs a function of judging whether a child is a pornographic child by combining information analyzed by the perceived property analyzing unit 220 and information analyzed by the soft biometric analyzing unit 230. That is, if the child pornography classification unit 240 determines that the image content is a pornographic material in the sensibility analysis unit 220 and the soft biometric analysis unit 230 determines that the image content includes a child, . ≪ / RTI >

The child pornography classifying unit 240 may include a search-based method for searching for similar images stored in a database serving as a reference of the child pornography, and comparing the information generated by the sensibility analyzing unit 220 and the soft biometric analysis unit 230 A method of classifying a sexually explicit sex image of a child into an input image after generating a classification model in advance using the feature information generated by the learning algorithm and the sensibility analysis unit 220 and the soft biometric analysis unit 240 So that child pornography can be classified in at least one image content.

2 shows an embodiment of the selectivity analyzing unit according to the present invention.

In order to analyze the sensibility of the image contents as described above, there is a method of generating and comparing hash values of the harmful video, a global characteristic of analyzing the sensibility by using characteristics such as the color, texture, Based sensory analysis method, a feature-based sensibility analysis method that detects feature points in an image and analyzes the sensibility using local area information around the feature points can be used.

The selectivity analyzing unit 220 of FIG. 2 shows a configuration for using the selectivity analysis method of the video content based on the global feature among the above selectivity analysis methods. 2 includes an image segmentation module 221, an image feature extraction module 222, a voice feature extraction module 223, and a selection degree classification module 234.

The image segmentation module 221 separates the image and audio from the image content, and transmits the image to the image feature extraction module 222. The voice is transmitted to the voice feature analysis module 223 ). At this time, the image segmentation module 221 may divide the image to be transmitted to the image feature extraction module 222 into a whole or a grid of a uniform size and transmit the segmented image.

The image feature extraction module 222 includes a plurality of different feature extraction units, such as a color feature extraction unit, a texture feature extraction unit, a morphological feature extraction unit, and a motion feature extraction unit, And extracts different feature information such as color, texture, shape, and motion in each area of the image.

The voice feature analysis module 223 extracts characteristics such as the repeatability of the voice pattern from the voice divided and transmitted by the video segmentation module 221.

The selection degree classification module 224 includes a selection degree classification model and a selection degree classification section. The selectivity classification model is created in advance through machine learning on feature information of previously collected images and voices. The selection degree classifying unit 224 compares the image feature information and the voice feature information acquired by the image feature extracting module 222 and the voice feature extracting module 223 with the selectivity classification model to classify the selected degree of the image content do. At this time, the degree of selectivity may be generated for each frame or analysis interval, and the degree of selectivity may be set to a value between 0 and 1. Here, 0 means general image and 1 means high selectivity. However, the value of the selectivity can be set to various values.

3 shows another embodiment of the selectivity analyzing unit according to the present invention.

Unlike the lexicality analysis unit of FIG. 2, which uses the method of analyzing the sensibility of the image contents based on the global characteristics of FIG. 3, the method is configured to use the method based on the local features. 3 includes a feature extraction module 225, a feature extraction module 226, a voice feature analysis module 227, and a selection degree classification module 228.

When the image content is received from the image extracting unit 210, the feature point extracting module 225 separates the image and voice, transmits the image to the image feature extracting module 222, and transmits the voice to the voice feature analyzing module 223 do. At this time, the feature point extraction module 225 extracts feature points of the image to be transmitted to the feature point presentation module 226 and transmits the extracted feature points, unlike the image segmentation module 221 of FIG. The technique of extracting feature points from an image has been proposed in various image processing techniques, so a detailed description thereof will be omitted here.

The feature point expression module 225 includes a plurality of different feature extraction units such as a brightness feature extraction unit, a color feature extraction unit, and a motion feature extraction unit. The feature point expression module 225 performs a feature extraction process on each feature point of an image , Brightness feature value, color feature value, and motion feature value.

Like the voice feature module 223 of FIG. 2, the voice feature analysis module 227 extracts characteristics such as the repeatability of the voice pattern from the voice that is divided and transmitted by the feature point extraction module 225.

The selection degree classification module 228 includes a selection degree classification model and a selection degree classification section in the same manner as the selection degree classification module 224 of FIG. The selectivity classification model is created in advance through machine learning on feature information of previously collected images and voices. The selection degree classifying unit 228 compares the feature point information and the voice feature information acquired by the feature point expression module 226 and the voice feature extraction module 228 with the generated selection degree classification model, Classify.

4 shows an embodiment of a soft biometric analysis unit according to the present invention.

4 includes a gender estimation module 231, a voice information analysis module 232, an age estimation module 233, an elongation estimation module 234, and a soft biometric information association module 235 .

The gender estimation module 231, the audio information analysis module 232, the age estimation module 233, the elongation estimation module 234 and the soft biometric information combination module 235 respectively extract the image contents And acquires information such as the sex, age, and height of the person included in the image. In particular, since the elongation estimation module 234 can not accurately measure the elongation of a person after the object having a limited size such as a face is detected, the elongation estimation module 234 estimates the average size of the predetermined object, the size of the detected object, The height information can be used to estimate the height. The gender estimation module 231 and the age estimation module 233 perform face detection on the image and then normalize the face using the feature points (eye, nose, mouth, etc.) of the face and then extract the feature information from the normalized face region And the gender and age of the person included in the image are determined through the previously learned gender classification model and the age classification model. In particular, the age estimation module 233 combines voice information, extension estimation information, and the like obtained from the voice information analysis module 232 and the extension estimation module 234 in addition to the information obtained directly from the image in estimating the age, You can estimate the age of the people involved. The gender estimation module 231 can also use the voice information acquired by the voice information analysis module 232 as additional information in addition to the information directly acquired from the image when estimating gender. The voice information analysis module 232 may be used to estimate the soft biometric information in the gender estimation module 231, the age estimation module 233, and the elongation estimation module 234, rather than directly acquiring the soft biometric information of a person included in the image content It can be provided as an auxiliary means for increasing the accuracy.

 The soft biometric information combination module 235 assigns an identifier for each person present in the image and the soft biometric information generated by the gender estimation module 231, the age estimation module 233, and the elongation estimation module 234, And performs matching to correspond to the identifier.

In the above description, the soft biometric analysis unit 230 includes a gender estimation module 231, a voice information analysis module 232, an age estimation module 233, and an extension estimation module 234 in order to acquire soft biometric information It is apparent that the soft biometric analysis unit 230 may include additional modules for obtaining more accurate soft biometric information. Since the gender estimation module 231, the voice information analysis module 232, the age estimation module 233 and the extension estimation module 234 are all soft biometric information estimation modules for estimating soft biometric information, Estimation module section.

FIG. 5 illustrates an example of application of soft biometric information analyzed by the soft biometric analysis unit of FIG.

5, the soft biometric analysis unit assigns IDs to at least one of the objects included in the image content as IDs, and outputs the IDs to the gender estimation module 231, the age estimation module 233, Sex, height, etc. of each of at least one person estimated in step 234 with an identifier.

The soft biometric information generated by the soft biometric analysis unit 230 may be generated for each frame or for each analysis interval and may include information such as sex, age, and height for a person present in the image content.

Fig. 6 shows an embodiment of a child pornography classification unit according to the present invention.

The child pornographic classification unit 240 includes a feature combination module 241 for generating a feature vector for classifying child pornography using the degree of selection and soft biometric information generated for each frame or analysis section, And a child pornography classification module 242 for judging whether or not the child is a pornographic child.

7 shows an example in which a feature combining module generates a feature vector.

FIG. 7 shows a case where the feature combining module 241 combines the selection and soft biometric information to generate a single feature vector. The feature for classifying child pornography can be generated as shown in Equation 1 by connecting each degree of selection and soft biometric information.

(Where F denotes a single feature vector, N _i (where i is a natural number less than or equal to k) denotes the selectivity, B _i denotes soft biometric information, and k (where k is a natural number) Indicates the number of frames.)

As described in Equation 1, the feature combining module 241 performs a feature extraction process in which the feature extraction unit 210 extracts the feature extraction result obtained from the feature extraction unit 220 for each analysis region or frame divided by the image extraction unit 210, ), And generates a single feature vector F for classifying child pornography.

The selection degree N _i may be represented by a real number having a value between 0 and 1 as described above and the soft biometric information Bi may be expressed by the sex, May include information processed by the total number of persons [male and female], the number of adults [male and female], the number of children [male and female] and the minimum age [male and female] have.

Fig. 8 shows another example in which the feature combination module generates the feature vector.

8 shows a case where the feature combining module 241 generates feature vectors separately for each of the selection and soft biometric information. In FIG. 8, the feature combining module 241 generates one feature vector (selectivity feature vector) for determining whether the obscenity is determined using only the degree of line extracted from the analysis section or the frame, For example, 0.7), i.e., a feature vector for classifying child pornography only for soft biometric information on an image whose feature vector is greater than or equal to a threshold value. A feature vector for determining whether a pornographic object is connected by connecting only the degree of line can be expressed by Equation (2).

(Where F _n denotes the feature vector and N _i (where i is a natural number less than or equal to k) denotes the selectivity, and k (where k is a natural number) denotes the number of analysis intervals or frames.

Then, the soft biometric information feature vector for the image whose feature vector F _n is equal to or greater than the threshold value can be expressed as Equation (3).

(Where F _b is a soft biometric information feature vector, and A _j (where j is a natural number less than or equal to m) is generated based on soft biometric information in an image having a feature vector F _n greater than or equal to a threshold value (M is a natural number less than or equal to k) means the number of statistical values. The statistical value is the average number of characters [male, female], average minimum age [male, female ], The rate of appearance of children [male and female], the rate of appearance of adults [male and female], and the rate at which both children and adults appear at the same time [male and female].

The feature combining module 241 may also generate the soft biometric information feature vector F _b regardless of the selectivity feature vector F _n . The soft biometric information feature vector generated by the feature combining module 241 using the soft biometric information extracted from the analysis period or frame irrespective of the selectivity feature vector F _n may be expressed by Equation 4 .

(Where F _b represents a soft biometric information feature vector, B _i (where i is a natural number less than or equal to k) denotes soft biometric information, and k (where k is a natural number) represents the number of analysis intervals or the number of frames .)

The soft biometric information (B _i ) is the number of persons [male and female], the number of adults [male and female], and the number of children based on information such as sex, age, Use information processed by [male, female], minimum age [male, female], and so on.

Fig. 9 shows an embodiment of a child pornography classification module according to the present invention.

9 shows an example of three child pornography classification modules 242. (a) shows a child pornography classification model 242 that uses a child pornography classification model generated through a machine learning method for a single feature vector to determine whether a child is a pornographic child Module 242, (b) a child pornography classification module for searching the child pornography reference DB for an image similar to the feature vector, and then determining whether the user is a pornographic child through a ratio of images including child pornography 242, and (c) includes N feature classification models corresponding to N feature classification sections and N feature classification sections, respectively. And N feature classification units generate N judgment results using the feature classification model for each feature vector, and then integrate the results to determine whether the child is a pornographic child.

FIG. 10 shows a method for detecting pornography of a child according to an embodiment of the present invention.

Referring to FIGS. 1 to 9, the child pornography detecting method of FIG. 10 will be described. First, the image extracting unit 210 extracts and acquires at least one image content stored in the image providing apparatus 100 (S110). Then, the extracted image content is analyzed and the image content is sampled (S120). The selectivity analyzing unit 220 analyzes the degree of selectivity of the sampled image (S130). Meanwhile, the soft biometrics information analysis unit 230 analyzes the soft biometrics information (S140). Then, the image content image extraction unit 210 determines whether the sampled image content is the end of the image content (S150). If the sampled image is the end of the corresponding image content, the feature combination module 241 of the child pornography classifier 240 generates the feature vector for the image content using the analyzed sensibility information and the soft biometric information (S160) . Here, the feature vector may be generated as a single feature vector in which the selection and soft biometrics information are integrated into one feature vector as described above, and the feature vector may be generated separately from the selection and soft biometrics information. Then, the child pornography class module 242 of the child pornography classifier 240 determines whether the image content is a child pornography item using the generated feature vector (S170). Here, the child pornography classification may be classified by discriminating whether or not the child is a pornographic child, but in some cases, a plurality of classes may be set in advance and classified into classes.

In the above description, the soft biometrics information is analyzed after analyzing the selected degree of the sampled image. However, the sensibility analyzing unit 220 and the soft biometrics information analyzing unit 230 may be configured to analyze the degree of selectivity analysis, May be configured to analyze information simultaneously in parallel. By analyzing the selection analysis and the soft biometric information simultaneously in parallel, the analysis time for the image contents can be reduced.

In addition, as described above, the selectivity analyzing unit 220 analyzes the degree of selectivity of the image content, generates a feature vector for the selectivity, and discriminates the corresponding image content from the generated selectivity feature vector as a selectivity image The soft biometric information analysis unit 230 may analyze the soft biometric information to generate the soft biometric information feature vector. In this case, since there is no need to analyze the soft biometric information unnecessarily for the image contents with low selectivity, the load on the system and the analysis time can be reduced.

11 shows a method for detecting child sexual pornography according to another embodiment of the present invention.

11, the image extracting unit 210 first extracts and acquires at least one image content stored in the image providing apparatus 100 (S210), and extracts the extracted image content The content is analyzed and the video content is sampled (S220). Then, the sensibility analyzing unit 220 analyzes the sampled image (S230), and the soft biometric information analyzing unit 230 analyzes the soft biometric information (S240). The child pornography detection method of FIG. 11 is different from that of FIG. 10, and it is determined whether the child pornographic material is the child pornographic material by using the selection and soft biometric information for the image in which the selection and soft biometric information are analyzed (S250). The child pornography determination information is accumulated and stored (S260).

(S270) whether or not the pornographic material of the corresponding image content is a child based on the cumulative stored child pornographic substance judgment information. If it is determined by the accumulated child pornography determination information that the video content is a child pornographic material, the child pornography is classified and terminated (S280). However, if it is determined that the child pornography is not the child pornographic material, it is determined whether the sampled and analyzed image is the end of the image content (S290). As a result of the discrimination, if the sampled and analyzed image is at the end of the image content, it is terminated without classifying it as a child pornography. On the other hand, if the sampled and analyzed image is not the end of the image content, the remaining image of the image content is sampled and the analysis is repeated (S120).

As a result, FIG. 11 does not perform the analysis until the end of the image content, and judges whether or not the child sexually explicit object is in advance for each of the sampled images, accumulates the child pornographic substance judgment information, and if the accumulated child sexually- , The remaining regions of the image contents are not analyzed further, but are classified as child pornography. Therefore, unnecessary analysis can be reduced to reduce waste of system resources and save time.

The method according to the present invention can be implemented as a computer-readable code on a computer-readable recording medium. A computer-readable recording medium includes all kinds of recording apparatuses in which data that can be read by a computer system is stored. Examples of the recording medium include a ROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like, and a carrier wave (for example, transmission via the Internet). The computer-readable recording medium may also be distributed over a networked computer system so that computer readable code can be stored and executed in a distributed manner.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is evident that many alternatives, modifications and variations will be apparent to those skilled in the art.

Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

Claims (1)

An image providing device capable of storing at least one image content and providing the image content to the outside; And
The method comprising: extracting at least one image content stored in the image providing device, analyzing the selectivity of the extracted image content to determine a degree of selectivity, determining softness of each of the objects included in the image content using a soft biometric technique, A child pornography determining device for analyzing the biometric information to determine whether the image content is child pornographic content; A child pornography detection system.

Images