Research On Filtering Images Technology For Clothing Search Engine

Recent years, shopping online has become a very important part of our life,especially shopping clothes. However, when people searching clothing by searching engines, appearing of junk clothing pictures not only wastes the user's time, but also increases the commercial search engines burden. If we can get a kind of picture garbage filtering technology to identify these junk images, the user will be able to get a better search experience, at the same time,the search engines can also improve the efficiency. Technology of filtering image which based on image feature provides a theoretical basis,however,the ability of different features make different. Therefore, we take local and global feature to improve the effectiveness of image filtering, the main contents are as follows:1)We analyze and study the existing domestic and international filtering images technology, and we pay more attention on content characterization and classification algorithms.2)This thesis presents a novel method to incorporate spatial information in the traditional bag-of-words model.The traditional bag-of-words model does not contain any information of the visual words in the image space.We take spatial information into consideration to improve the representation. Experiments on our data show that, our method provides a better result of recognition rates.3)In this thesis, a method of sliding window features extraction principal component images are presented. The traditional method of principal component analysis method (Principal Component Analysis, PCA),usually make an one-dimensional image vector to analysis, it will lose a lot of information. In our framework,we present a sliding window to represent one image. By this way, we take advantage of the sliding window to retaining edge structure. Experimental results show that filtering image recognition accuracy with this method is superior to traditional algorithm.