Research On Image Recognition Based On Pre-classification

Image recognition is a technology in the field of computer vision, which using digital image processing technology and pattern recognition method to recognize interest targets in the image. This technology has been widely used in synthetic aperture radar image recognition, map navigation, license plate target detection, and medical lesion diagnostic and so on.In real life, images are often influenced by external factors, which making images produce the problems of deformation, noise and blur. That increases the difficulty of image recognition. At the same time, the scale of image library is growing, the content of images become more and more rich, and current image recognition system often has the shortcomings of slow recognition speed and its recognition accuracy needs to be improved. Therefore, this paper proposed an image recognition method based on pre-classification, this method can improve the efficiency of image recognition system while gaining high recognition accuracy. The main research work of this paper as follows:① For the problems that computation time is long and recognition accuracy is not high which exist in image recognition process of large image recognition system, this paper proposed a pre-classification idea. That is, through pre-classify the given template image library before identifying the target image improving the operating efficiency of the image recognition system by narrowing the identification scope.② Chose the scale-invariant feature transform(SIFT) algorithm to extract image features after comparing several image feature extraction algorithm, because it has strong robustness to variety of changes in the external environment, contains a wealth of information and has good scalability. Introducing the bag of words(Bo W) model and using visual vocabulary vectors to express image features which can greatly reduce the complexity of feature similarity calculation. Taking into account of the fact that support vector machine(SVM) has good generalization ability and global optimality when solve the problems such as nonlinear, over-fitting and high dimension pattern recognition, using SVM to classify. Based on the above points, this paper studied an image classification method based on SIFT and Bo W, uses for the pre-classification process of the image recognition method based on pre-classification which was proposed by this paper. The simulation results in the standard image library show that this method has good classification effect.③ Considering an image matching and recognition method based on SIFT and the random sample consensus(RANSAC) algorithm, used for the recognition process of the image recognition method based on pre-classification method which was proposed by this paper. Using RANSAC algorithm, based on the SIFT matching algorithm, to eliminate mismatched feature points in the matching process. The experimental results on the local affine-invariant image datasets show that this method has strong robustness.④ Giving the design framework and specific implementation process of the image recognition system based on pre-classification which was proposed by this paper. Doing recognition simulation experiments on the large-scale image library using the open source software packages of LIBSVM in MATLAB. The experimental results show that, the method proposed by this paper can greatly improve the efficiency of image recognition while get high recognition accuracy.