Modified variance-based transition region extraction and thresholding

Variance, a common statistical measure reflecting degree of deviations between mean and individuals, could be used to describe the degree of gray level changes. Hence, it can be utilized to depict and extract image transition region. Unfortunately, variance is sensitive to salt & pepper noise. To solve this problem, a modified variance-based transition region extraction and thresholding method is presented in this paper. The proposed method performs statistic analysis on image local window to exclude salt & pepper noise, improving accuracy of transition region extraction. In addition, we also develop a scheme for ruling out this type of noise in image thresholding result, further enhancing its noise immunity for salt & pepper noise. The proposed algorithm was compared with famous local entropy-based method and gray level difference-based one on a variety of images, and experimental results show its effectiveness and noise immunity for salt & pepper noise.