Image Retrieval Based on Fuzzy Kernel Clustering and Invariant Moments
Pengyi Hao, Youdong Ding, Yuchun Fang · 2008
KFCM is a new clustering method, which is applied in pattern classification in some way. However, it has not been applied in the field of content-based image retrieval (CBIR). Considering the predefinition of clustering number and membership matrix is still a hot potato, a novel predefinition approach is proposed in this paper, what's more, a new CBIR method is suggested and validated. First, values of images are extracted, then they are clustered by fuzzy kernel clustering, and edges are detected by Canny operator, finally, edge invariant moments are calculated and formalized. After that, the Euclidean distance between two images' formalized moment vectors gives a measure of similarity. Experimental results show that the proposed algorithm having better ability in the anti-noise, and the precision and recall are remarkably improved as well.