Svm Ancient Chinese Character Image Recognition Based On Feature Fusion

Images of ancient chinese character recognition is an important research field of pattern recognition, in view of the ancient chinese character recognition process with multiple process, multiple constraints, etc.,and processes are interrelated between the various processes may more easy fail to local optimum, leading to identify inefficient. The Support Vector Machine (SVM) can flexible to decide boundary in a high-dimensional feature space, because of its strong global convergence. Therefore, this paper, based on SVM method, using coarse classification and fine classification of the two-layered classification and recognition ideas to identify the ancient Chinese character image.On the basis of full analyse the graphemic of ancient chinese characters and the feature of the input image, in view of ancient chinese characters for different input images may be caused by differences in pretreatment effects, with adaptive functions proposed preprocessing algorithm, which shields the uncertainty preconditioning effect caused by different input image, To a certain extent ensure the effectiveness of the pretreatment effect, and also to better eliminate to drag-in the fasle feature in the pre-processing stage, lay the foundation for the feature extraction.As the requirements of high-quality feature extraction can't be satisfied by a single characteristic of structural or statistics feature.A new method was proposed in this paper which is the integration of multi-feature fusion extraction method between structural features and statistical characteristics.That is the serial integration feature extraction methods between structural characteristics of components and the global dot density characteristics and the parallel integration feature extraction method between grid-based features and characteristics of the local point density. These methods are good to meet a high degree of differentiation, high stability, and typical of the feature extraction requirements.Two-layered based on SVM classification model be established by Extracted high-quality information, which is a coarse classification model with multi-feature serial integration and a Fine classification model with multi-feature parallel integration. The results show that the classification model based on hierarchical thinking which is proposed in this paper has higher precision.In this paper, the ancient Chinese character recognition image classification system is developed. The results show that the system has high classification ability.The methods which proposed in this paper about image recognition of ancient Chinese characters is provided a new feasible method for the current study few areas of ancient Chinese character recognition.