Adaptive Thresholding Based Image Segmentation with Uneven Lighting Condition

Abstract

We propose two new schemes for segmentation of images with uneven lighting conditions. These are based on adaptive window selection. The first one is a window merging method based on Lorentz information measure (LIM) but the second one is a window growing method using the notion of entropy. We propose two new window merging criterion where the window merging is carried out based on linear combination of local and global statistics. In window growing method, we define a notion of feature entropy and the window is selected employing jointly entropy and feature entropy. The two window merging schemes perform better than the schemes using only LIM. The proposed window growing technique is compared with schemes using only LIM and the proposed two merging techniques. It is found that window growing technique is best among all in the context of error due to misclassification error.