Shape Matching Algorithm Study And Application

The LED automatic wire bonder is a modern high-tech microelectronic packa-ging equipment which synthesizes technologies of precision machinery, automatic control, image recognition, optical and ultrasonic welding, and it is mainly used to weld wires that connect the chip pad and outside framework. The image posit-ioning system is one of key technologies to weld automatically, and difficulty of the system is high demand of welding timely and positioning accurately.Scholars have proposed various fast positioning methods so far, and the most common is template matching; but traditional template matching algorithms whic-h base on the gray value have the problems of complexity and high sensation on illumination, noise and rotation. It is very difficult to meet increasing applied demands.To solve above problems, the thesis proposes a new shape matching algorith-m based on edges, combining with images characteristics, designs and implements the algorithm. It is proved that the algorithm can meet the requirements of indu-strial production by testing. The paper mainly completes the work as follow:Firstly, it introduces traditional template matching algorithms, such as normal-ized cross correlation, sequential similarity detection algorithm, and analyze the adaptation under noise, illumination, occlusion and small angle rotation and the corresponding solutions. The results show that these algorithms can't meet requi-rements of image recognition system of automatic LED wire bonders, and briefly illustrate the feature-based template matching algorithm.Secondly, the dissertation proposes and implements a shape-based template matching algorithm. First of all, extracts the edge points of template image usin-g Canny operator, and correct manually, then using the Sobel operator filters im-ages. The results are the template image position vector and the gradient vector of edges; that is the template shape model. Similarity measurement uses the inn-er product of the template edge points and the corresponding points of image s- earch region. Analysis shows that the method resists disturbing from non-linear illumination, noise, occlusion and not sensitive to small angle rotation. On the search strategy, the thesis proposes adaptive regional search method which is ba-sed on the probability model. Searching region is further divided into large and small probable areas. The algorithm cuts the search area, so it greatly improve the searching efficiency. It also gives the corresponding solutions for the conditi-ons of early termination and sub-pixel accuracy.Finally, the paper analyzes the hardware and software of the LED automatic wire bonders. The Object-oriented approach is used to construct a class of shape-based template matching and tests the algorithm for the actual image of LED aut-omatic wire bonders. Get the algorithm performance data in case of noise, illumi-nation, occlusion and small-angle rotation. The data shows that the shape-based t-emplate matching algorithm basically meet the requirements of real-time and posi-tioning accuracy, and robust to noise, not sensitive to light, suitable for small an-gle changes.