A Study On Image Recognition Algorithm Base On Template Matching

As an elementary and important technology of computer vision, image matching is widely used in navigation, industrial inspection, remote sensing image analysis etc. For industrial application, image matching technology with high precision, strong reliability and real-time ability is the aim that people try hard to in pursuit of all the time. This paper is dedicated to the algorithm research of image matching applied to industrial inspection. After analyzing and comparing several matching algorithms, a new matching algorithm that bases on edge characteristic and hierarchical strategy is presented and proved to be reasonable by experiments.The basic concept of image matching is first introduced in this paper, including mathematics description, the category of matching methods, the key problem in image matching, the performance demands of image matching and its influence factor. Then some image preprocessing methods that relate to the matching algorithms in this paper are introduced, including Gaussian filter, geometrical transformation, edge detection etc. These preprocessing methods hold the balance to improve the performance of matching algorithms.Secondly, several template matching algorithms are discussed from the aspect of character space and similarity measure and some improvement is presented respectively. A fast method of calculating geometric moments is presented in the matching algorithm that based on centralized moments. A heuristic criterion is presented in the matching algorithm that based on edge characteristic. These improved methods are proved to be reasonable and effective by experiment and compared with the method using image pixel values directly. Experimental result indicates that the improved matching algorithm based on edge characteristic can greatly enhance matching speed without reducing matching precision and has good performance under the influence of several error factors such as image noise, illumination change, contrast change etc.After the foundational research of taking edge characteristic as the similarity measure, edge pyramidal structure and hierarchical search strategy are used to achieve real-time matching performance. This improved algorithm can also have the capability of detecting rotated angle and is proved to be effective by experiment.Finally, a computer vision software of locating which has the function of template matching is designed by Matrox Image Library (MIL) combined with the apply of image capture card and motion control card.