Image Retrieval Based On Color And Non-Subsampled Contourilet Features

With the rapid development of information technology, huge amounts of digital images and videos continue to emerge. In order to effectively use these resources, content-based image retrieval (CBIR) technology is proposed, and quickly becomes a hot research field. Images'low-level features are extracted mainly in this technology, and their similarity is measured. How to extract and use images'low-level features for fast and effective retrieval is the key technology in this field, and also is the main problem to be solved.The low-level features of the images include color, texture, shapes and spatial relationships. In this paper the color and texture features are focused on to do deep analysis and research, then the histogram and the non-subsampling contourlet transform theory are combined, and an image retrieval method based on salient region's color histogram and non-subsampling contourlet texture features is proposed. The main research work and innovations are as follows: Firstly, the salient points are detected and the image's annular salient region is positioned, then the histograms from the image's salient and background region are created and combined with the weighted method, so the salient region color histgram proposed by this paper is got. This approach can not only get the image salient region histogram flexibly with spatial color information, but also ensure the rotation invariant. This method overcomes the commonly used color features'disadvantage, can extract image features effectively and improve image search results obviously. Secondly, we study the non-subsampled contourlet transform theory, and propose a non-downsampling contourlet and entropy method to describe the texture. This method can decompose the image to get a strong direction and ensure the translation invariance, effectively reduce the dimensions of texture features, and strengthen the image retrieval's advantage. Thirdly, the color and texture features are integrated, and a new method based on salient region's color histogram and non-subsampling contourlet texture information is presented. The combination of these two types of features, descripting respectively the global and local characteristics of images, and goes over each other's defects. Finally, we construct an image retrieval model that integrates color and texture features; a retrieval system is created with the model, and it implements the methods that are discussed above; we use every method to do experiment. The results show that the proposed method has better retrieval performance and higher efficiency than method based on single type of features and the traditional comprehensive feature method.