Research Of Image Recognition Based On Neural Network

The Image recognition is developing in the last 20 years, which has been becoming the main contest of the classification and description for some objects or process (regarding as the Image), so the image recognition has many extensive researches. However the traditional image recognition technique is mostly based on the foundation of large-scale calculating, such as the image recognition of statistics and the Sentence construction, but there are too much contradiction between the amount of calculation and the accuracy of recognition. In recently, the new solution for this problem has been researched following the developing of the neural network. The technology of image recognition based on Neural Network is a kind of new-style image recognition technology, which is developed with the theory of present computer technology,image processing,artificial intelligence and pattern recognition. It is a method of the image recognition with the algorithm of the Neural Network based on the traditional image recognition.This thesis analyses the current image recognition methods based on the Neural Network. According to the image recognition characteristics, two image recognition models was put forward, they are the combining of the genetic algorithm and the BP Network, and the Support vector machine. And give two models'learning algorithm and detail applied technique. The main task of this thesis is in the following of aspects:1.The thesis concisely introduces the theory of image recognition and some kinds of method commonly used in pattern recognition.2.The thesis sets analysis on the introduction of feed-forward Neural Network model and algorithm. It combines together the genetic algorithm with BP Network, solving the question that the network of BP is easily to get bogged down in the partial dinky weakness.3.The thesis sets emphasis the models of support vector machine which can get the best answer when the sample is not enough. We analysis it's more mode classifying question and putting up a new strategy into it. A kind of dynamic sample training method is putting forward, which could make SVM add new kinds of recognition class anytime you want.At last, arming at practical matters in image recognition, the thesis describes the application of the car plate recognition system by GA-BP and the face recognition system by SVM is shown .