Multi-subgraph matching for logo localization using genetic algorithm
Pengfei Wang, Yali Zheng, Hong Cheng, Nan Run Zhou · 2016
Given a query graph model, most of existing graph matching algorithms focus on finding a one-to-one mapping in a target graph. However, there are multiple objects in the target image, such as logo retrieval. In this paper, we propose an algorithm, called multi-subgraph matching, for Logo localization. We try to recover an exact one-to-k mapping, which directly constrains multi-objects corresponding to one query logo in the model. Essentially, multi-subgraph matching is a combinatorial problem, so we solve our problem in the genetic algorithm framework. Assignment vector is coded as chromosomes from selection, crossover to mutation to obtain a global optimum. We test our algorithm on the real data for logo localization, and compare with one state of the art - spectral matching (SM). Our algorithm shows better results than SM on the data.