Mixture Feature Based Junk Image Filtering

Text based image search tools such as Google and Baidu will cause lots of junk image, because only text near the image was searched. We construct a content based junk image filter system to resolve this problem. We use kernel based KNN to clustered the image searched by the Google into multiple classes by its vision feature, then use hyperbolic technique to display the image sampled from the clustered image, when user selected the image wanted, save the images which class equal to the image selected and filter out others.