Histogram statistics based variance controlled adaptive threshold in anisotropic diffusion for low contrast image enhancement

It is very difficult in low contrast images to distinguish between the noisy background and the regions of low gray level inter-region edges. The classical Perona–Malik anisotropic diffusion is able to smooth the defective background but cannot enhance faultless low gray level inter-region edges in such low contrast images. The proposed method provides an unsupervised machine learning process to modify the anisotropic diffusion by generating an adaptive threshold in diffusion coefficient function using statistical measures. In the proposed method, image histogram is employed to calculate the global gray level variance over the entire image and local gray level variance over the defined neighborhood of each pixel of given image. The adaptive threshold in diffusion coefficient function varies in accordance with the difference between the two variances which gives a measure of intensity contrast in that neighborhood. The experimental results from various low contrast images have shown that the proposed unsupervised machine learning approach for adaptive threshold selection in anisotropic diffusion can effectively smooth noisy background with preservation of low gradient edges.

► An adaptive machine learning approach for low contrast image enhancement.
► Histogram based global variance computation over the entire image.
► Histogram based local variance computation over predefined neighborhood.
► The adaptive diffusion coefficient function using difference between two variances.
► The diffusion results demonstrate the enhancement of low gray level edges.