The integrated annular region image retrieval technology cumulative histogram and co-occurrence matrix
|Course||Signal and Information Processing|
|Keywords||Image retrieval Ring segmentation Accumulative histogram Gray level co-occurrence matrix|
With the rapid development of multimedia technology and Internet and the increasing requirement for image information, the content-based image retrieval are emerging and developing. Color and texture features of visual characteristics are widely used in content-based image retrieval. Therefore, a study has been made in this paper on color and texture method and a retrieval method aiming at the shortcomings of the existing methods is proposed based on the color and texture features.The main research work and contributions of this thesis are listed as follows.(1) A color-feature-based retrieval method using ring segmentation is proposed. Firstly, the image is partitioned into ring areas. Then, the color characteristics are extracted from the segmented ring areas to be fused with image space information, thus avoiding the sensitivity to image rotating and size changing, etc. when using the fixed block segmentation method. Finally, the accumulative histogram is extracted to effectively avoid the zero values flaw. In this paper, the Euclidean distance arithmetic is employed in measuring the similarity of color features, which highlights the areas of representing the important content of the image.(2) A texture-feature-based retrieval method is proposed in this paper, which presents an improved gray level co-occurrence matrix fused with Tamura features. Firstly, the proposed algorithm counts the probability density of gray difference between two neighboring pixels instead of the joint probability density. In addition, the statistics of the gray difference value is taken in more directions around the pixel point, and a smaller value of the distance between two neighboring pixels to extract the texture information with more details. Finally, via combining some Tamura texture features, the ultimately extracted texture features are more consistent with the human eye perception. In this paper, the ratio method arithmetic is utilized to measure the similarity distance of texture features.(3) Due to the unstably retrieval result of the retrieval system with a single feature, a retrieval method combining the accumulative histogram of ring region and gray level co-occurrence matrix is proposed through normalizing the color and texture features(4) The retrieval system is designed for performing the proposed algorithms above in this paper. The experimental results have shown that the retrieval precision of the proposed color-feature-based method and the proposed texture-feature-based method outperform the fixed block method and the traditional co-occurrence matrix method, respectively. Furthermore, the retrieval results using the proposed combination color and texture features method are more precise than the single feature method.