Research On Image Segmentation Algorithm For Machine Vision System Of Die Bonder

Die bonder is one of the key equipments in chip assembly line, related to electronic, optical, mechanical technology, is a high-speed, high precision micro-electrical mechanical system. Machine vision system is among the most important parts in Die bonder. To date, machine vision abroad has been developed greatly and produced important applications in fields such as industrial automation, robotics, and satellite observation of Earth. But in China, researches on machine vision are just started a few years ago and most of the existing vision products are bought from other countries. So it is necessary to study on machine vision thoroughly.For machine vision systems, captured images are often degraded to various extents because of the lens spherical aberration. In order to solve this problem, this paper introduces two new segmentation methods: Image segmentation method based on BP neural network and optical degradation model.Image segmentation method based on BP neural network. BP neural network has the capacity of parallel computing, distributed saving, self-studying, fault-to-learnt and nonlinear function approximating. So it's widely used in the classification area. When applied to image segmentation, the pixels in the image are clustered to object and background. So the object we are concerned about can be extracted from the image.Image segmentation method based on optical degradation model. Aiming at the problem that the image with non-uniform illumination can not be efficiently segmented, image segmentation method based on optical degradation model is proposed. The algorithm first constructs the optical degradation model, then uses the degradation model to correct the illumination of the original image and finally employs global threshold to make better segmentation. Two experiments are designed to demonstrate the proposed algorithm, and qualitative comparison is made among Otsu's segmentation, adaptive thresholding approach and the proposed algorithm.By evaluating with experiments on Person Computer, the results show that the two proposed segmentation algorithms are more accurate and efficient than the usual thresholding segmentation .