Research On Content-based Image Retrieval

This paper mainly focuses on the study of content-based image retrieval, and proposes some rotational invariant textures, which are the structure descriptor of motifs. Moreover, two content-based image retrievals related with motifs are produced. The specific researching contents are as follows:Firstly, depth research and analysis of content-based image retrieval of relevant search technologies, which are image retrieval based on color-feature, image retrieval based on texture-feature, and image retrieval based on shape-feature. What’s more, the detailed description of the content-based image retrieval process, namely feature extraction of query image, pre-processing of image dataset, similarity comparison and feedback results. Lastly,the effects of content-based image retrieval evaluation are depicted specifically, that is,precision and recall, the mean precision and F-measure.Secondly, 24 kinds of the image texture features that is motif are extracted in 2 * 2 grids with priority. Meanwhile, in order to achieve the rotation of the image information retrieval,the original 24 kinds of motifs are further improved, and summed up 8 kinds of rotation invariant motifs.Finally, the paper proposes two image retrievals, i.e. an image retrieval based on motif co-occurrence matrix from the angle of texture space distribution and an image retrieval based on motif matrix from the point combined by color-feature with texture-feature. The image retrieval based on motif co-occurrence matrix takes advantage of eight motifs to extract the feature of texture in gray image and the motif co-occurrence matrix, made up of row motif co-occurrence matrix and column motif co-occurrence matrix, is considered to be a feature matrix. The image retrieval based on motif matrix combines the color feature of average pixels with the texture feature of motif, and proposes a motif matrix which counts the frequency of the average and the corresponding motif as the feature matrix. Lastly, the paper takes advantage of Matlab, Java and My SQL to develop the two above image retrieval methods, and performs contrast experiments in Corel image dataset and Brodatz texture dataset.