Study Of Adults Picture Identification Based On Contents

With the speedy development in information techniques and Internets, The network information becomes a kind of convenient information source and recreational way that be familiar with. But large numbers of smuttiness information and erotica has already seriously interfered the normal network living, harmed the teenager's mind and the body's health. The spread of erotica and smuttiness information has caused the concern in the world. How to purify the network environment, increase supervises and control the means to the network activity, improving the ability that information identify has became a kind of strong need. As its techniques supports, bad information identification techniques based on content have caused people's recognition increasingly. According to the adult's picture of the contents identify and examine, the techniques recently have caused people biggest interest. It is a research lesson in network percolation system base on the contents, an importance for facing and need the solution at the same time. In fact, the adults picture identification is a image classification. We research in this area adopting the covariance classification method to identify the adult's picture. The key techniques: Skin Detection, Target District Pick up, Picture Character Pick up, The Design of the Classification Machine.The skin color detection is placed in the important position in adults picture identification based on contents. It is the foundation of character pick up to process in design with classified method in statistics. We studied the skin character, color-space chooses in process of skin detection and compared different skin color models. We use YUV and YIQ color-space to detected skin district. And in this district, we use OTSU to pick up the skin again. The experiment shows that it can examine the skin district effectively. According to the characteristic of fails detected skin district and the true skin district, we use the method of veins analysis to eliminate the mistake in skin detection.Character Pick up is the foundation of the method in classification with statistics. How to choose and choose which appropriate characteristics is the key to identify the bad picture. We described the main methods of characteristic pick up in picture; include veins analysis and border shape characteristics and region shape characteristics etc. We noticed that the shape of the target district can be described by its pixel distribute in the different circumference in its quality heart. We present a scheme based on center distributing of the region to descript shape feature and use it to pick up the shape characteristics in the targets district. Combine Fourier transform to pick up the low frequency picture characteristic of the bad picture district, we do further benefit to the bad picture identification.The design of classification machine is the key to the method of classification in statistics. The problem of bad picture identification is a small sample problem. Support Vector Machines (SVM) has its excellent learning performance and has a very big advantage in small sample classification, not linear classification and high dimension mode classification. After introducing the theory of SVM, we described the construction of eigenvector, the choice of kernel function and its method of training, the rudimental framework of the bad picture identified method based on SVM. The experiment shows that the entire rate of this project is 82.67%. It has got a good result and come to the study target that we expect. The study of this project can provide a technologic aegis to cut erotica picture's spread.