Research On Image Matching With Slight Deformation

Image matching plays an important role and achieves hot research topics in many fields, such as digital signal processing, pattern recognition, artificial intelligent, machine learning, meanwhile, they are also widely used for image retrieval, brand detection, robot automatic localization, aircraft visual navigation, scene understanding. Its main goal is to realize specifically cognitive functions of human being. In this paper research, it involves three main study topics of image matching with slight deformation, which are object shape modeling, shape feature extraction and representation, shape matching based on optimal searching strategy. This paper achieves three main contributions as follow:(1) A new biologically inspired shape model based on visual recognition theory is presented. This shape modeling approach is inspired by the visual recognition principle of HMAX bionic model, which performs more robust, widely adaptive and better to fit these properties of biological visual recognition.(2) An effective approach for extracting and representing shape hierarchical visual features is provided. In this paper, there are three logical layers of shape visual features as: the visual point, the visual line and the visual region, which are distinctively and robustly extracted with multi-scale space analysis and visual context technology. Meanwhile, these shape visual features are sparsely selected with important properties of visual recognition. This paper approach shows to be more robust, complex and efficient.(3) An efficient hierarchical matching strategy based on dynamic programming table for shape features corresponding is achieved. These shape features corresponding are efficiently carried out with hierarchically matching strategy, which combines global dynamic programming approach and local Bipartite Graph Matching method together. As many tests show, this paper method is more robust, efficient and achieving better recognition performance.