Commodity Image Retrieval Based On Content

With the popularity of the internet, electronic commerce is entering a new age. The image information of goods as the main reference of the user selects merchandise information plays an important role in electronic commerce. How to choose the right commodity in the large numbers of goods quickly and accurately has very important application value. Content based item image retrieval is becoming a new direction of item image retrieval in the electronic commerce system, this paper has done some research about this, the main content include:(1) At first, study the image retrieval technology, analysis the framework of image retrieval system, some common image global feature extraction method, feature vector similarity measure method, normalized characteristic normalization and image retrieval performance evaluation method.(2) This paper study using the local features of the images to do item image retrieval, besides, according to the characteristics of local feature, the bag of visual word model and document inverted index structure is applied to speed up the retrieval.(3) Considering the SURF feature is extracted on the gray scale image and ignored the color information of images, this paper proposed a new method about combining the local feature and the color layout feature to do the retrieval. And then some experiments have been done to prove the effectiveness of this method.(4) Aiming to reduce the effects of complex background in product images, this paper proposed processing methods based on image segmentation. For the query image with complex background, using interactive segmentation method called GrabCut, this method only need to draw a rectangle around the product on the query image, and then it can extract the main product accurately. For the image in product image library, this paper presented a method based on JSEG segmentation algorithm. According to the experiment result, this paper’s method makes the correctly rate of searching results has been improved.