Content-based Image Regional Skin Feature Research

With the development of computer network technology, it has brought dramatically change to human's life and work and it is the main means that human obtain information from network. However, the network with mixed messages in addition to good news, it is also filled with a lot of bad information, such as pornographic images, which do great harm to the growing man and human's healthy life. So relevant departments under the State determines to remediate network t, this method will give back a healthy network environment to user by handing out the files such as"a network of green dam"document. The paper takes the features of the image skin to research, which can be used to filter the pornographic images with important theoretical and practical significance. This article did a series of researches based on the above background. As pornographic images have the common feature: a lot of bare skin. This paper did researches on human skin to get a series of features by experiments and theories in order to provide technical support for establishment of green dam.First of all, describes skin color detection principle and related technologies in detail and experiments, and ultimately chose the color space detection model and texture detection model. This paper uses wide traversal algorithm and sets the threshold. Results show that the proposed method can detect skin regions of human. Firstly, the paper did researches to the skin detection principle and related technique and chooses the color space detection model and texture detection model by the compare of experiments. The experimental results show that the method can well detect the image of human skin areas using the breadth traversal algorithm and setting the threshold denoising method. Secondly, binary image can be formed by using the color space model; while the value of the point will be set 1 if meets the condition and 0 if not. The external rectangle of human in image is done by getting the upper left and bottom right point coordinates after denoising, which can separate the human body area and background effectively and complete image segmentation. Thirdly, this paper comes out a new idea that makes the external rectangle into several internal rectangles after studying lots of pornographic images and finding out that there is lots of exposed skin, this paper analyzes the various regions of the distribution of exposed skin to separate the pornographic images and non-pornographic images. Finally, integration of multiple features form a multi-Eigen value vector and normalization processed; the vector is used to filter the pornographic images and detect pornographic images with the example of similar images in order to achieve the purpose of filtering.The method of features is tested by Kapoix search platform, and results show that algorithm can retrieve similar images with the case diagram, with high accuracy (Precision) and detection rate (Recall). The technology can be extended in Image filtering for the network, such as e-mail attachment filtering the image.