The Research On The Technique Of Image Pattern Recognition Based On Hough Transform And Moment Invariants

Image processing based on pattern recognition is one of the most importantresearch directions in the image recognition field with the computer technologydevelopment.The main content of this paper is the application of Hough transform andmoment invariants theory in pattern recognition. In this article, patternrecognition is used as theory foundation and image process technology as basictechnology. Combining with the new theory tools (neural networks), this paperpresents effective resolve methods to three parts of image recognition (imagesegmentation, image feature extraction and image classification) throughsystemic analysis of it.The dissertation is organized as follows:At first, the main methods of pattern recognition in the image processing,which include statistical decision, structural method, fuzzy method, artificialneural network are summarized. The general strategy of pattern recognition andthe key technologies are also analyzed. At last, the system structure of patternrecognition is concluded.To solve the pattern recognition problems, the neural network can achievethe involved demarcate of earmark space, and it cut out for high accelerationand parallel systems to achieve. The neural network technology and featureabstraction method is combined to recognition the picture in this thesis, and theBP and RBF neural networks are adopted. Did some simple and involvedexperiments on picture recognition.In this paper, the technology of automatic characters recognition and Houghtransform are researched and explored. A method of weight integration isproposed and it is presented a new method for characters recognition accordingto the property of characters' structure by using the duality of point-line inHough transform. In this method, we use the number of the points which are onthe same line and its corresponding coordinate in a character's parametric spaceas the feature input vector of RBF neural network recognition system. To dothis will reduce the dimensions of the feature vector greatly and will reflect thegeneral distribution of the character's dot matrix and the essential feature of itsfiguration in the round. Hough Transform is applied to parameters selection andthe adaption of the number and position of data centers of RBF neural networksthat are realized in this method. Consequently, RBF neural networks designedwith this method can generalize well.The moment technique is widely used due to its appealing mathematicalsimplicity and versatility, such as image processing, computer vision andpattern recognition. According to the basic principle of image segmentation, apre-processing algorithm is showed, which is fit for this thesis. Aim at thetarget which has been translated, scaled and rotated, a recognition methodbased on improved order-normalized moment invariants is improved.