A new image recognition algorithm based on skeleton

Traditional recognition methods which mainly match object images with their skeleton couldnpsilat resolve well complex objectspsila recognition problems. So in the paper, with an introduction and improvement of moment invariants, a new image recognition method is proposed with the combination of skeleton and moment invariants. Firstly, the paper analyses the thoughts of method. Then, the concept of object main skeleton and its extraction method is described, and with view to the characteristics of the skeleton, an extended Hu moment invariants algorithm is brought forward to calculate moment invariants of the skeleton. At the recognition stage, a two-layer generalized regression radial basis (RBF) neural network is adopted to do machine self-learning and target-identifying. Compared with the present recognition methods based on similarity matching with skeleton, the algorithm doesnpsilat need to face many problems such as the difficulties in matching and realizing based on skeleton graph, the complexity of the shock graphs, the object selectivity of the Reeb graphs and the order of the nodes which canpsilat be guaranteed in SA tree and so on. Compared with traditional moment recognition methods, the method not only can make calculation results meet scale, translation and rotation invariance, but also can reduce the number of related efficient pixels during moment calculation. In the meanwhile, it overcomes the difficulties that traditional moment recognition methods encountered when they deal with the fuzzy object boundary, and thus is effective. Finally, some experiments prove that the algorithm has better results for general object recognition.