Image Retrieval Based On Combined Features

In this thesis, the research background, the important of the issue, the up-to-date applications and the development of the key techniques of content-based image retrieval are reviewed. First, we concentrate on the research of image retrieval using color, texture and shape information. Then, we discuss the region-based image retrieval methods from different aspects including image segmentation, image region representation and similarity measurements. Region based image retrieval is a better method which simulates the human comprehension of images, and hence the semantic gap is narrowed. The proposed method is proved its validity and efficiency though theoretical derivation and experimental results.The main research work and innovation of this thesis are as follows,1) To represent spatial color features, we proposed an image retrieval method based on the spatial distribution of local colors. The spatial color distribution information is obtained by dynamic sub-block splitting method. The image is split into sub-blocks according to the size of the object in the query image, and the weights of the sub-blocks can be adjusted.2) Content-based image retrieval using color and shape is studied. To represent the shape content of an image, the edge orientation autocorrelogram is used. To represent color features, local color cumulative histogram is computed, and to solve the problem of lacking the spatial knowledge of the histograms, we also extract the color moments of the partitions. The Guassian model is used to normalize the feature distance. Combining the above three normalized distance, the global similarity measure is obtained. Better image retrieval performance can be achieved by using the combined features.3) A new feature vector is proposed by combining color, texture and position information. An adaptive image segmentation method is proposed based on the finite mixture model and the EM algorithm. The Gaussian mixture model and t mixture model are used. The modified SMEM and the Greedy EM algorithm estimate the parameters of the mixture models. Images are segmented according to the Bayes...