Image-based Object Recognition Algorithm Research

Image-based item identification technology is a hot issue in the field of computer vision. Accuracy of object detection and recognition is a difficult point of current research.The main problem in the image-based feature point is: how to select efficient image feature point to solve the problems of image translation, image rotation, scale variation, at the same time reduce the effect of block, image noise in the process of items recognition, in order to achieve better accuracy of goods identification. In view of the problems, this thesis do research on items identification technology based on feature point.Based on analysis of the existing image matching technology, this thesis concentrate on feature point matching technology that is SIFT(Scale Invariant Feature Transform). Analysis of SIFT algorithm is from four aspects: establishment of DOG(Difference-of-Gaussian) scale space, using space extreme points to detect, descriptor generation, and similarity measure based on k-d-tree. For the disadvantage of SIFT algorithm, this thesis improve it. Experimental results show that SIFT algorithm has better stability in terms of influencing image matching such as illumination change, image scale transformation, rotation transformation, image noise and so on.In order to improve computation speed of the algorithm, this thesis do study and realize feature point items recognition technology that is SURF(Speed UP Robust Features) algorithm. The idea of SURF algorithm is that it compute quickly based on integral image, establish scale space using square filter, detect image extreme point using fast-Hessian matrix and produce the descriptors of feature points. On account of the wrong match problem in the process of applying SURF algorithm to image matching, this thesis come up with RANSAC algorithm to purify characteristic point to resolve efficiently this problem. This article do detailed comparative analysis of performance of this two algorithm by comparing the experimental data of SIFT with SURF. Experimental results show that SURF algorithm has improved tremendously at the speed of items recognition based on characteristic point. Finally this thesis analog and realize the system of items recognition based on image on the computer.