Automated ear segmentation in various illumination conditions

Biometrics systems based on ear are still in need of more investigation to make it robust and accurate. The most critical step in ear recognition is segmentation as all subsequent steps will depend on the accuracy of segmentation. In this paper, a robust ear segmentation method is suggested. The proposed method consists of a sequence of steps. First, a Biased Normalized Cuts method is applied to initiate the ear image segmentation process. Then, the segmentation process is perfected by performing the following functions: gray-level slicing, entropy, thresholding, skeletonization, image filling and opening. Finally, a substitution process is applied. Our proposed algorithms are tested on ear images captured under various illumination conditions and produced encouraging result. A 95 percent accuracy rate is achieved at an average of 10 seconds processing time.