Research On Interactive Clothing Image Search And System Implement

Recent years, with the continuous development of the Internet, people have begun to buy from the store commodities to buy goods through the network, but the network on a wide variety of goods and the number is too difficult for users to quickly find the products they need. How to quickly find the match for the goods they need from the vast amounts of commodities has become a people need to be resolved. Some Internet companies to provide users find approximate product image retrieval services to meet the needs of the user’s search, but the text description of the product image information is incorrect and the image background noise factors make the commodity image retrieval service can’t meet the user’s actual demand. And interactive retrieval also problems to be solved is how to multiple target objects in the same image. In order to solve these problems and improve the performance of the commodity image retrieval, in this paper, clothes, bags and other goods with complex background image an interactive product image retrieval method. The main contents and contributions are as follows:First, Interactive image segmentation algorithm, the algorithm take full advantage of the user to choose to be extracted location information of the rectangular area where the image of goods, The main idea is to assume that the closer the location of the pixels in the image from the center of the rectangle, this pixel as a foreground of the image the greater the probability, on the contrary, this pixel greater the probability of the image background. And spatial location information the GrabCut algorithm based on the combination of Gaussian mixture model modified by color alone, Thereby improving the accuracy of the foreground of the image extracted.Secondly, unsupervised product image segmentation algorithm, the algorithm uses the face detection method to determine the target may exist a region to be extracted; simple analysis of the area of the target object exists, and to determine whether this detection region having reasonable. If the target area detection is correct on the image background and the foreground of image were constructed Gaussian mixture model; Gibbs energy function iterative minimum cut energy reaches the approximate minimum and image segmentation. If the target area detection error or no human face is detected, using main object extraction algorithm to extract the commodity target.Third, the interactive product image retrieval method, first, we use of the goods of unsupervised image segmentation algorithm to filter images in the background noise in the back-end database, and filter the product image after image background extract color features and texture features were constructed database of feature; then we use of interactive image segmentation algorithm to extract the target object to be retrieved commodity and extract color feature, texture feature. Finally, under feature matching calculated commodity target search results.Fourth, in this paper, I have completed an interactive product image retrieval system. Image database contains two sets of image data of a total of2.4million of the original image and the segmented image. The system can be able to achieve real-time interactive image retrieval of goods on a single computer.