The Research Of Image Retrieval Technology Based On Visual Attention Model

With the rapid growth of digital image quantity, how can efficiently and quickly retrieve the image data in a lot of image information which it has become important problem in the current image applications. The first, the content of images will be described in order to search images, and to establish the image index through represented formal image content. Due to the complexity of the image and the subjectivity of human cognition, the efficient and general image index is a very difficult task. At present, the visual content index can generally be described by low-level visual features (or physical characteristics). The semantic content index can be described by semantic characteristics (it can be described by the text), in addition to it can take the man-machine interactive way, and the semantic gap will be overcome in the future.It is still very difficult that the image semantic features are extracted with the existing technology. Therefore, the further development of CBIR is very difficult, such as the enormous gap between the lower image visual features and high-level semantic, namely the image semantic understanding of user can not full expressed by the image processing algorithm of the low-level visual features extraction. In order to overcome the semantic gap, an objective visual attention model is constructed in this paper, it is based on the object of attention and interest region detection method, wherein an detection method of automatic extraction the region of interest is proposed for image. On the one hand, the regional interest measure is efficient used by this method in the visual attention mechanism. The other hand, the area of actual physical meaning can be gained by this model in image. This visual feature extraction mainly includes three parts in model.(1) The salient point feature extraction, it is measured by local salient points and global salient points. The local salient points are determined by dynamic threshold value method.(2) The mark descriptor feature extraction, the operator according to the internal and external markers based on watershed segmentation, using evolutionary programming method to extract mark descriptor features.(3) The regional feature extraction, it segments images using evolutionary programming method, and the object region is generated. As the candidate area is selected by attention focus, so the most significant area is target region of interest in image.Finally, the three characteristics are combined as a symbol in image features, the similarity. Results show that this method is well to the image retrieval results.