Research On Content-based Image Retrieval Based On Wavelet And Visual Perception Texture Features

Recently, with the development of multimedia technology and Internet, the volume of digital image is becoming bigger and bigger. The image has important application in military and civil domains. How to retrieve image effectively has became a hot research topic.Traditional image query technology by text (QBT) depends on manual labeled image. However, QBT has distinct shortcomings: different people has different perception on the same image content, and manual labeling leads to that it is impossible to work with the large amount image data in the Internet. In 1990's, content-based image retrieval (CBIR) was proposed. Content-based image retrieval uses the image itself, such as color, shape and texture feature, instead of any description of the image as the query condition. The retrieval system extracts the query image's features, and searches in system data sets. Content-based image retrieval has been extensively investigated by more and more researcher.In the paper, we reseach on content-based image retrieval based on wavelet and visual perception texture features. First, we study the problem of reducing the dimension of feature vector for texture image retrieval based on wavelet decomposition. The reduction technique is used to overcome unnecessary computation burden and even inhance the retrieval performance. Second, this paper proposes some new texture featuresbased on wavelet transform' subbands which coincide with human eyes perception. Third, we analysis local features for image retrieval. In order to extract the salient points more exactly, we propose an improved salient points detector. The experimental results demonstrate that the method is effective.