블록 질감 히스토로그램을 이용한 내용기반 영상 검색 시스템
- 원문 URL
Due to the widespread communication networks and the technical development on multimedia, the multimedia data such as still images and moving pictures are increasing rapidly. Therefore, in order to effectively manage the multimedia data, the classifying and searching techniques are required. We proposed the content-based image retrieval method using a new texture feature, Block Texture Histogram. Previous GLCM based methods use the texture features from all over an image for the image retrieval. So, they have the disadvantage not to use local information of images. When two images have the same distribution of texture, but the one has objects of local textures and the other has the general background of textures to be scattered all over it. It is not easy to distinguish the one from the other by the previous methods. However, our proposed the retrieval method can solve this problem. At first, we segment an image into N×M size blocks and generate a GLCM per block. Then, the statistical texture features like Contrast, Energy, Entropy and Uniformity, are calculated from each GLCM. The Texture Feature Map is constructed from the statistical texture features. At last, the Block Texture Histogram is generated by the Texture Feature Map. To evaluate the performance of the proposed method, we used natural images, as test images and calculated ARR and ANMRR, as the performance measures. The experimental result showed that the proposed method had the improvement of 15.4% for ARR and the improvement of 16.2% for ANMRR, when compared with the previous methods. In conclusion, the proposed method shows the higher performance of the retrieval recall and precision than the previous methods. Especially, we confirm that it is able to retrieve effectively the image with some objects which have textures to be gathered locally.