Shape retrieval and recognition based on fuzzy histogram

• Fuzzy histogram is introduced for building shape descriptors.
• Fuzzy shape context and inner-distance fuzzy shape context are proposed based on a 2-D uniform fuzzy partition.
• Locally constrained matching is proposed for computing shape distance.
• Our method demonstrates effective performance compared with others algorithms.

Shape representation and shape matching are significant topics in computer and human vision. In this paper, fuzzy histogram model with a uniform fuzzy partition is introduced instead of the classical histogram model, and two fuzzy-histogram-based descriptors are proposed: fuzzy shape context (FSC) and inner-distance fuzzy shape context (IDFSC). Compared with classical-histogram-based descriptors, FSC and IDFSC provide more accurate descriptions of samples distributions in log-polar space. Based on fuzzy-histogram-based descriptors, a novel shape matching framework named locally constrained matching (LCM) is proposed for computing the dissimilarity between shapes, and the rotation invariant problem of descriptors can be properly settled. Experimental results on a variety of shape databases show that shape retrieval and recognition results can be effectively achieved by using the proposed method.