The Research Of High-performance Template Matching Algorithm In Machine Vision

Machine vision is an essential component of industrial processes as diverse as semiconductor manufacturing, automobile manufacturing, and pharmaceutical packing. It is a subfield of engineering that consists of computer science, mechanical engineering, optics and industrial automation. The image processing is the most important component, which consist of template matching, edge detection, optical character recognition (OCR), and so on. Template matching can deal with element localization and recognition. In order to promote the domestic automation equipments, it is necessary to research and develop a high-performance machine vision algorithm.In this thesis, we research and develop a fast, high accuracy and robust template matching algorithm based on image gray values. Since normalized cross correlation (NCC) is invariant to illumination changes which will often occur in virtual application systems and is robust to noise, it is a very good similarity measure method. In order to make the matching for real-time, an innovative hierarchical search strategy is proposed, which can dramatically reduce the number of computations. With the purpose of searching the rotation images, we propose a new rotation invariant matching algorithm, which select the inscribed circle data of different rotation template as the matching region. To obtain the location of the object with higher accuracy, we use an algorithm that combines similarity resampling with surface fitting, then we extract the local maximum or minimum of the fitting polynomial analytically.The template matching algorithm was implemented by VC++6.0. Experiment results have proved that it is a real-time, subpixel accuracy and rotation invariant algorithm, and it can be simply carried out, which can meet the industry automation equipment needs.