Research On Image Recognition Algorithm Based On Deep Belief Network

Handwritten Chinese character recognition technology has become one of the research hotspots in the field of image pattern recognition,which can make people's daily life more convenient,and has a broad market space.In recent years,it has received extensive attention and research in the academic community.The traditional handwritten character recognition system mainly includes three parts:image preprocessing,feature extraction and feature classification.The pros and cons of feature classification directly affect the recognition accuracy,which in turn determines the performance of handwritten Chinese character recognition technology.In recent years,driven by the high-speed computing performance of computers and the massive data of the Internet,deep neural networks with highly abstract feature extraction and classification capabilities have been widely used in image recognition and computer vision,and a large number of breakthroughs have emerged.Inspired by this,this thesis studies the handwritten Chinese character recognition technology based on deep belief network.The specific work is as follows:Image preprocessing and feature extraction of Chinese character image.The preprocessing process of Chinese character image mainly includes image acquisition,noise removal and binarization scheme design.Furthermore,the noise of handwritten Chinese character image is effectively removed by wavelet transform,and it is carried out by Gauss local adaptive method.In the process of feature extraction of the Han image,we use wavelet transform to extract the edge features of the pre-processed image.Handwritten character classification and recognition.Several common algorithms of feature matching are introduced particularly.After analysis and comparison,deep belief network is used as the main body of the classifier.By analyzing the structural characteristics of deep belief network,the shortcomingsof its activation function are searched out.Wavelet basis function is used as the activation function of deep belief network,and the selection of activation function is also compared.Wavelet basis as activation function can improve the recognition accuracy of deep belief network,especially for the slight change of Chinese character information.Particle swarm optimization(PSO)optimizes classifier.carefully analyses the learning process of deep belief network,explains the shortcomings of traditional algorithms,introduce both PSO and its feasibility as well as builds network model and scheme design.The comparative experiments explain that this paper can effectively improve the recognition rate and speed of handwritten character recognition by using the wavelet deep belief network to recognize handwritten character image offline,and to achieve good results.