基于改进模板匹配的晶圆划切算法

Algorithm for Wafer Slicing Based on Improved Template Matching

摘要: 当前国内晶圆自动划切的工业应用中，主要采用基于灰度的模板匹配方法，但其计算量大、划切效率较低。本文提出一种改进的基于几何边缘的模板匹配算法，通过Canny边缘检测生成合适的边缘模板，计算模板边缘曲线的梯度方向作为匹配信息进行相似度计算，采用相似度阈值判断优化搜索策略，然后在图像金字塔最顶层进行粗遍历匹配获得潜在匹配目标，逐层匹配直到最底层，以达到对传统的几何模版匹配算法进行加速的目的。实验表明，该方法具有较高的鲁棒性，对于目标均匀或非均匀变化的光照、部分遮挡的情况可以得到良好的匹配效果，而且算法在保证精度的同时可以满足实时性要求，适用于晶圆图像自动划切的工业应用。

Abstract: Nowadays the domestic industrial applications of wafer automatic slicing mainly adopt gray-based template matching method. However, its calculation is quite time-consuming with low slicing efficiency. This paper proposes an improved template matching algorithm based on geometric edge to achieve the goal of accelerating the traditional algorithm. It generates appropriate edge template by Canny edge detection; calculates the gradient direction of template edge curve, which is used to calculate similarity as matching information; optimizes the searching strategy by using the similarity threshold determination. Then, the rough traversal matching is implemented at the top layer of the image pyramid and the matching process continues layer by layer until the bottom. The result of experiment shows that the algorithm proposed in this paper performs high robustness, which can obtain good matching result under different conditions of objectives, including uniform or non-uniform illumination and partial occlusion. Besides, it meets the real-time requirement while the accuracy is ensured, which can be applied to practical industry of automatic wafer image slicing.