정규화 레이블링 불변모멘트에 기반한 다중객체 영상검색

Abstract

In CBIR with multi-objects, It shows an interest in the image retrieval method based on object features. In this paper, we have proposed an image retrieval method based on normalized labelling features of multi-objects. First, it extracts, normalizes and rearranges the object-labels which present image shape information. Secondly, it measures image similarity using moment features invariant to transition, rotation and scaling of the objects. The simulation results using Multi-object database showed that our proposed method is superior to the edge histogram method and the wavelet shape extraction method. The experimental results demonstrate the high retrieval efficiency i.e Recall 0.23, Precision 0.85. Consequently, our retrieval system shows better performance than the conventional system that cannot express the shape of objects exactly.