Segmentation of die patterns using minimum cross entropy

The application of a minimum cross entropy thresholding algorithm to die pattern segmentation is presented. A combinatorial derivation is given which shows that this method maximizes the probability of a random experiment which generates the image data using its segmented version as a model. The algorithm is computationally efficient and results are in good agreement with the principle of maximum entropy in providing an unbiased estimate of the image. The algorithm was applied to the image segmentation of a die pattern in a wire bonding machine and was found to be superior in terms of computational requirement and robustness to the change in light intensity