Research and Applications of Key Techniques on Image Filter
|School||Beijing University of Technology|
|Keywords||Sensitive image filtering Skin color model The statistical color space model Texture model Support Vector Machine|
How to prevent the network pornography infringement is an important technical issues . Just blocked URLs and filter pornographic text messages means that web filtering is not enough , they must bring the computer vision technology to combat pornography and dissemination of the images embedded in the home page . In this background , relying on the \a sensitive image filtering architecture , experimental proof is based on the architecture of the system effectively have to filter sensitive image . Construct an experiment Gallery is the basis of all the work in this article , the authors first construct a gallery , then future research work carried out on the basis of this gallery . Skin color model as the core of the architecture of the label Gallery on training in to produce color model and the use of the mask image of the area of the skin color model of the original image processing produce marked complexion . However , simply using a color model can not accurately identify areas of the skin , observed that the smooth skin areas connected region , and therefore on the basis of the previous step of the introduction of the texture model for the further processing of the original image to produce marked skin areas the mask image . In order to better filter , and use on skin color mask images and skin mask image extracting characteristics classifier filter image classification . In color processing stage of work is experimental and a new color model - Statistics color space model . In addition , the authors also proposed on how to measure the color model separability of the training set , and a fast Gaussian mixture model parameter estimation algorithm . Main in texture processing stages of the work is to select on the basis of a study of previous work the texture model suitable for the system . The main work in the processing stage of the classifier structure and experiments prove in favor of classification filtering feature vector classification algorithm selected for the system on the basis of the study of previous work .