出版社：The Institute of Image Information and Television Engineers
摘要：We proposed a new image retrieval algorithm with partial indexing capability, which emphasized the local features of images by combining the salient points in a wavelet domain. These points representing sharp parts of images such as edges were detected. Then using a spatial orientation tree structure, they were combined into the Low-Low ( LL ) frequency subband to emphasize the local properties of the image. The wavelet coefficients in the modified LL subband and their color moments were then saved as image feature vectors, which both preserved the necessary image content and yet had enough discriminating power for image retrieval. The experimental results demonstrated that our image retrieval algorithm significantly improved retrieval accuracy, computational cost, and storage space of feature vectors in a general image database, and that partial retrieval performed much better in retrieval accuracy and robustness.
