Image Retrieval Based On Low-level Features

With the development of computer science and technology, the application of Internet and the rapid increment of multimedia database, we have a lot of digital images and video information. In order to manage and retrieve those information, the CBIR(Content-Based Image Retrieval) has emerged to be one of the hot research areas in image domain.In this dissertation, lots of exploratory research work has been done around some key techniques of CBIR. The main contributions of this dissertation are summarized as follows:(1) A new image retrieval method based on the partition of color edge is proposed. Firstly, the color edge of 24 bits ture color image is extracted by using canny detection operator. Secondly, the color edge image is divided to some grid regions, and the color and texture histograms for each grid are computed as image features. Experimental results show that the proposed image retrieval is more accurate and efficient in retrieving the user-interested images.(2) A robust content-based color image retrieval using multiple features is proposed, which not only takes into consideration the important image edges, but also utilizes the support vector regression (SVR) theory. Firstly, the image denoising and color edge detection are performed by using the SVR classification. Secondly, the whole color edge is divided into some local grids, and the color histograms and texture histograms for local grids are computed as image features. Finally, the similarity between color images is computed by using a combined feature index based on the color histogram and texture histogram for local grids. Experimental experiments, including comparisons with state-of-the-arts, show the effectiveness of our algorithm in improving the retrieval performance (Especially for noisy images).(3) A color image retrieval approach based on Pseudo-Zernike moments and color feature is proposed, which not only takes into consideration the important color information, but also utilizes the Pseudo-Zernike moments theory. Experimental experiments, including comparisons with state-of-the-arts, show the effectiveness of our algorithm in improving the retrieval performance.