The Pornographic Image Detect Based On Compressed Domain

The poison of pornography is in flood the serious influence young people physical and moral integrity, simultaneously also gave the people to use the Internet to bring many inconveniences normally. How prevents the poison of pornography the violation is the important research subject, had already attracted the domestic and foreign numerous enterprises and the development facility is engaged in this aspect the research. In the home, the Chinese counter-yellow software from the end of last centry, until now, the multiform counter-yellow software already had the nearly 20 kinds, these filtration software realizes the technology which often uses is the website blockade and the sensitive key word match. But along with the Internet scale and the information unceasing inflation, uses this technical existence obvious hysteresis quality and the limitation, must therefore unify the image filtration technology to be able more effective to prevent the network poison of pornography the dissemination.Along with the technical unceasing development, Internet has become the information important source. While the network is widely used, massive pornographic image also by illegal dissemination. Although the people already took certain measure to carry on the filtration to the network sensitive image, but because exists in the network the picture majority is by the JPEG compression form existence, the traditional sensitive image detection algorithm must decompress the first compression image to the spatial domain is carrying on processing and the recognition, thus increased the computation complexity, has limited the recognition system performance enhancement. Carries on the sensitive image directly on the image compression territory the examination, has saved the time which the image transforms, may enhance the sensitive image detection the examination speed. Carries on the sensitive image on the compression territory the examination, because its speed is quick, the computation complexity is low, is suitable for the real-time examination the request. Therefore the utilization becomes the topic which based on the compression territory's sensitive image filtration technology urgent.This article is precisely in above thought foundation, first in the compression territory progress pedestrian face examination and the skin color examination, then unifies the person face examination and the skin color examination test result synthesis withdraws the characteristic value, sends in the decision tree sorter, thus establishes a sensitive image detection model, realizes carries on the recognition to the sensitive image the image filtration system. This sensitive image detection system can achieve certain examining rate. And compares with the former method, because this system has saved the complex IDCT operation, from and raised the sensitive image detection examination speed.This article depended on 2004 year Zhuhai technological project (PC20041101) - -"based on the content sensitive picture filtration technology research and in IE browser's realization", introduced how method which unified using the compression territory's on person face examination and the skin color examination information, based on this and withdrew the sensitive image the recognition clues, finally has constructed a network sensitive image filtration system. Specifically speaking, this article main research work includes:1) On compressed domain of image feature extraction is through the excavation at the time of image compression between the results of the final code stream or the information contained in, in pursuit of the decoder does not decode or part of the case of extracting the characteristics of image content characterization in order to achieve image The purpose of treatment. Based on compressed domain of image feature extraction, the ideal situation is a direct stream on the final feature extraction. On images compressed domain feature extraction techniques generally can be divided into two categories: transform domain-based and space-based technologies. Traditional face recognition methods, in the face of the compressed images to identify people, the first image to the "all-extracting" operator, and then to identify, this is undoubtedly a major constraint on the identification of system performance improve. To this end this paper, a compressed domain on effective methods of face recognition.2) Color detection can be used as face recognition and facial expressions of people the initial extraction steps. At the other person face tracking and gesture-tracking system, color detection has played a role in positioning the initial goal, to determine the exact color region or not, the whole system directly determines the identification, detection accuracy. In the sensitive image filtering system, the body's skin color is the most important characteristics. Sensitive to the issue of classification of images largely depend on the measurement accuracy of color. Color detection is therefore sensitive to distinguish between the basic picture.3) Feature selection is very important to the classification of its serious impact on the design and performance. The first step in the process of image recognition is the first analysis of the effectiveness of the various features and select the most representative characteristics of implementation of the system This is where the difficulty. In this paper, the feature extraction is based on the basis of skin color detection, we mainly based on the model generated by the skin color image mask to extract the features of. In the field of image retrieval, the index commonly used features of color, texture, and some low-level and a simple shape characteristics. With easy calculation of these characteristics, the characteristics of stable performance. Image retrieval can be applied to the characteristics of a mechanism sensitive to the image in the detection and filtering. Taking into account the most sensitive images have more exposed skin, and through many experiments and analysis, we can extract a series of good character. Characteristics of our classification of color in color characteristics of the majority, taking into account the shape of the characteristics of color information block.4) Then, the traditional algorithm and our proposed new method, to build a multi-level classification system for sensitive images. Out of time and efficiency point of view, we select a decision tree classifier to verify the above-mentioned characteristics. Characteristic which withdraws in the person face examination and in the skin color test result foundation, uses a simple decision tree sorter, has realized a multi-level sensitive image filtration system. Stemming from simple and the efficiency consideration, this article only utilized the simple classified tree to carry on the examination to above characteristic, but has not conducted the careful research to the sorter. But in the classified examination experiment may see, this article proposed the person face and the skin color characteristic in the sensitive image classification's application is feasible effective. These characteristics may use for to provide to the more complex sorter use.5) Studies in the process, we were analyzing unceasingly summarize in the predecessor research results foundation to construct a sensitive image filtration system frame gradually, has realized the sensitive homepage online measuring ability well. And, based on the content image filtration system is the core, but the image filtration is also the expansion which carries on the skin color examination module foundation and extends.The experimental results show that compared with the previous method, the image-sensitive detection system to effectively improve the image of testing the speed-sensitive, while at the same time must guarantee the accuracy of detection.