Research On Color Image Retrieval Based On Visual Attention

In nowadays, with the advancement of the science and technology and the development of the information,such as the development of multimedia and the application of Internet. A great deal of digital images are produced everyday which include many different fields, no matter public or military. Face with the mass image database, we need to analyse and manage those information reasonably, and also may retrieve at any time and any where. The traditional image retrieval can not satisfy the requirement. In order to retrieve the information of users need quickly and effectively, CBIR(Content Based Image Retrieval) has emerged as the times require and become a hot spot of information field.In this dissertation, a lot of exploratory research work has been done around some key techniques of CBIR. The main contributions of this dissertation are summarized as follows:(1)In order to advance the performance of image retrieval system, and the unstabitily only using single feature, we propose a color image retrieval based on multiple features of image edges. Firstly, we use the Canny detection operator to extract color edge. Secondly, color histogram and direction histogram about the extracted color edge image were computed as the image features. Finally, we calculate the similarity between color images by using the combined feature index based on two kinds of histograms. Experimental results show that the proposed method is more accurate and efficient in retrieving the user-interested images.(2) According to the bit-plane theory and the noise attack characteristic, a new robust color image retrieval based on significant bit-plane is proposed. Firstly, the significant bit-planes are extracted from the color image. Then, the Harris-Laplace detector is utilized to extracted interest points from the significant bit-planes. Finally, calculate the fuzzy color histogram based on salient points. Experimental results indicate that the proposed image retrieval outperforms other schemes in terms of retrieval accuracy and stability. Especially, it can retrieve the noise (including fuzzy, and illumination, etc) image effectively.(3) The existing color histogram has many problems, such as lost of spatial information and the wide gap between low-level visual features and high-level semantic, and it can not distinguish the difference between the different images. According to the human visual system(HVS), a new image retrieval method based local visual attention is proposed in this paper. Firstly, we extract the feature points by using the SURF method. Then, enlarge the feature points. When we get the color features, we firstly calculate the color complexity measure(CCM) of feature points. Secondly, all pixels are weighed by the color complexity measure, and the weighted histogram are cumulated. Finally, the similarity between color images is computed by using the weighted color histogram. We use first order central moment, second order central moment and third order central moment of the spatial distribution entropy to get the texture features. At last, we use the color and texture features to get the image retrieval. Experimental results show that the method is flexible and accurate to describe the feature of an image and increase the image retrieval precision and recall.