Study On Matching Algorithm Of Weld Image Processing

To track and process the weld image information fast and effectively is the key technologyduring the process of automatic welding. As an important branch of the digital imageprocessing, image matching technology is widely used in target tracking and object recognition.To improve the matching precision and reduce the complexity of time effectively, thispaper mainly studies the template matching algorithm and stereo matching algorithm. Putforward a kind of "rough matching-fine matching" matching model, we get left weld image andright weld image through two cameras respectively and extract the son figures which have thesame size as the template image using template matching algorithm and then calculate the densedisparity map using stereo matching algorithm.This paper proposes a template matching algorithm based on mutual information. To makeup for the deficiencies of high computation complexity, we propose the image segmentationthreshold value iteration algorithm based on the maximum entropy by using the imagethreshold segmentation technology and iterative thought which regards the information of thetarget area in the image as segmentation standards and save these information as much aspossible. Experimental results show that our algorithm enhance the matching speed andprecision effectively.Then we propose the stereo matching algorithm based on the belief propagation theory. Thetraditional belief propagation algorithm based on pixel exist two shortcomings: large amount ofcomputation and a single pixel may cause matching error easily. This paper proposes a novelbelief propagation to make up for this two shortcomings. The algorithm is based on imagesegmentation and adaptive mechanism which turns stereo matching problem into energyminimization problem and get the dense disparity map by the reference image colorsegmentation and the adaptive match point. Experimental results give the support to thealgorithm.