Research On Image Retrieval Based On Coding

In recent years, with the development of internet technology, the wide use of handheld devices, and appearance and rise of social network service, the images are produced at a more and more high speed. However, how to retrieve the target images in an image database with millions of images quickly and precisely becomes a challenging issue in information technology. With the appearance of some new excellent local feature descriptors which convey lots of image information and new models of image representation, some new Content-Based Image Retrieval methods appears. Compared to the traditional complex and inefficient Text-Based Image Retrieval, these new methods show great advantages and they become the focus of researchers gradually. In this paper, we mainly focus on achieving the code expression of an image based on the features and Bag of Words(Bo W)model. The main achievements of this paper are as follows:1. A new feature descriptor called edge orientation gray difference which contains a lot of information of an image is proposed and a combined coding method for image retrieval is proposed. To achieve more information of an image efficiently, we find a way to combine our descriptor and the Scale Invariant Feature Transform(SIFT) descriptor while encoding images. The experiment results show our descriptor has an ability to express the information of image and our combined coding method leads to a high precision and recall.2. A combined feature coding method by SIFT and Local Binary Pattern(LBP) feature of superpixels in this paper is proposed. Firstly, each superpixel is encoded in both feature space respectively and by combining the codes in different feature space, a new combined code of a superpixel is achieved; Finally we calculate the similarities between images by codes of images. This method helps us train the codebook which has a large size and encode the images quickly.3. A new coding method based on the constraints of inter-image is proposed. During searching the process, images are encoded by the constraints of query image and accomplish an image retrieval method. The experiment results show our coding method has a high precision.