Study On Image Retrieval Based On Visual Attention

The number of digital images has been increasing dramatically in recent years and as a result acrisis is now taking place in many fields that the image content are required heavily. Many fieldsand industries including telecommunications, entertainment, medicine, and surveillance, need highperformance retrieval systems to work efficiently. This trend will grow quickly as wireless networkdevelops.Since image feature extraction methods are used basically for some features of each imageitself, ignoring the relationship with other images. Salient regions in images are often the mostattentive. The classic visual attention models are usually based on pixels or features to build salientmap, but not to analysis the whole image.Considering the shortages mentioned above, we do some works as follows:1. An image retrieval method based on spatial color histogram of oriented graph is proposed,in which it used color,edge,space and others intuitive image features. Experiments validate theeffectiveness of the proposed.2.An image retrieval method based on feature clustering is proposed. Firstly, RGB values areextracted and preprocessed, and used in K-means clustering. And then sparse coding for each patchand pooling for the feature vectors are performed, in which the feature vectors are normalized.Experiments show that the proposed has good characteristic of less time consuming and high meanaverage precision in image retrieval.3.An image retrieval method based on visual attention is proposed. Firstly, salient regions aredetected using low-rank representation, and features represented by over-complete dictionary basedon feature clustering are extracted. Then feature vectors are reduced in dimension and representedby structure sparsity, which is successful in image retrieval from experiments.