Multivalent Graph Matching for Symbol Recognition

Kieu Diem Ho, Jean-Yves Ramel, Nicolas Monmarché

Abstract: This paper deals with a symbol recognition problem where symbols are deformed in different ways. To address this difficulty, the multivalent graph matching problem is considered to allow more matching possibilities between the symbols. More precisely, a solution to the multivalent graph matching problem is formulated as an extended graph edit distance minimum with additional splitting and merging operations. This minimization problem is tackled with a variant of ant colony optimization, the max-min ant system. Besides, a neighborhood search strategy added to the solution building process of the max-min ant system aims to accelerate the computational time. The efficiency of the proposed approach is illustrated in a symbol dataset in several aspects, covering both quality and quantity analyses. The result shows the interest in using multivalent graph matching to deal with noisy symbols and their meaningful interpretability in the context of sub-part correspondence against other bijective graph matching-based approaches.