The Design And Implementation Of Photo-sensitive Recognition System Based On Prohibited Gallery

With the rapid development of Internet technology, almost companies from home and abroad pay more and more attention on e-commerce. Images in e-commerce becomes more widespread used to display of goods , the images become increasingly important, but some of goods displayed in image will appear some prohibited pictures, which serious harm to the e-commerce Environmental and social harmony.To solve the above problems, Alibaba, as the largest e-commerce platform in the world, developed photo-sensitive recognition system, which based on the prohibitive image gallery that depended on the OpenCV. The system used to determine whether the pictures that customers'upload for goods, is prohibit or not.This paper describes the traditional method of image retrieval and content-based image retrieval method, and then discuss image feature extraction methods, which based on the feature point detection, especially introduce the SIFT feature point extraction method, and then according to SIFT algorithm's advantages and disadvantages and image characteristics we made some modify to SIFT algorithm, which aims to reduce redundant features and improve the matching accuracy. Firstly using SIFT algorithm extracted the feature point of pictures which stored in prohibited picture gallery, and then put the feature points that extracted by SIFT algorithm into the database. We have pictures'prohibited feature points extracted from database. In accordance with certain laws, the system generate a prohibited image feature tree, through the use of prohibit pictures feature tree system can quickly find the ratio of similar between the images which the network provider and prohibit pictures greatly, can reduced the matching time and improve the efficiency of matching. According to the result score which generates in the process of matching between the images which the network provider and prohibit pictures'features tree. Information of the image will flow into manual area to audit or direct delete the picture's information in the database, or do nothing to the pictures.