Salient Region Detection With Application To Content-based Image Retrieval

With the development of digital media technology, more and more digital images are produced. How to organize, store, represent query and retrieval these images is an urgent problem. In the scenario of image retrieval, since the background usually occupies large area of the image, the image feature extraction is negatively affected. In this paper, we propose a method that extracts salient region before image retrieval, which is more favorable to content-based image retrieval.Based on the Visual Attention Model, the thesis studies the salient region extraction technologies and image retrieval technologies. The main research results are as follows: (1) Salient region extraction. This method uses a visual attention model to extract focuses of attention in an image and makes them as interest points. The algorithm decomposes an image into three features: intensity, color and orientation. For each feature we use Multi-resolution Gaussian Pyramid Decomposition Method, and thus obtain a series of images. Then we compute saliency map for each feature by using Center-Surround Calculator and combine these maps to obtain the general saliency map. Then an algorithm based on the Gaussian Mixture Model is used for image segmentation. By combining the saliency map and the segmentation result together, we can compute the salient region of images. (2) Content-Based Image Retrieval: We discuss the methods of image retrieval based on salient region, and compare the retrieval performances between this method and the conventional image retrieval algorithm.By comparing the differences between the proposed image retrieval method and the conventional global based method, we can draw the conclusion that the former can improve the retrieval accuracy. Several discussions are given in the end.