Image registration is a fundamental task in image processing used to match two or more pictures taken, for example, at different times, from different sensors, or from different viewpoints. The first part of the thesis is concentrated on the problem of two layer IC micro- images registration. Variations of intensities between the two images which caused by eroding is the most difficulty in image registration. We analyze the merits and disadvantage of several classic registration algorithms through experiments. A two steps algorithm is designed to match image from coarse to fine scale. First, phase correlation method is used to get coarse result on large scope, then correlation coefficient method is used to get precise result. Both of them are implemented by FFT. The method is last, accurate, high noise-robust and have a good result in IC micro-image registration. It has theoretical and practical value in other image registration problems. IC wire net recognition and auto-trace based on two layer images is another important part of the thesis. An integral approach for structure unit recognition based on IC knowledge constraints is developed. Using abundance information of two layer images and the information around the wire hole, the track operator and the discrimination operator is designed to recognise the wire hole. Experimental results indicate that our integral recognition approach accords with the work of experts and has special advantages. Image mosaicing is to "stitching" together overlapping multiple images into a large image. One of the difficulties in image mosaicing is, when the overlapping region has not evident features or has many similar features, pseudo-match will happen and cause mosaicing failure. Based on MSE curve and constraints between the images, an approach was proposed for distinguishing and solving of pseudo-match. Experiment results demonstrate the efficiency of this method. As the independence of this part, it was attached as appendix at the back of the thesis.
修改评论