When the basic distribution constituting one gray level histogram is presented as a non-Gaussian distribution

such as gamma

extreme value

Rayleigh

uniform or beta distribution

how to automatically select the best possible segmentation threshold is still quite challenging. To deal with the issue of threshold selection in the above-mentioned different gray level distributions

we propose an automatic method of threshold selection that is guided by maximizing cross-region Shannon entropy under edge guidance and contour constraints. This method utilizes constant guiding edges and dynamically changing contours to construct a series of continuously changing one-dimensional gray level histograms

and adopts Shannon entropy as the entropy calculation model. Therefore

it can calculate the cross-region Shannon entropy across several local regions in the image

and it takes the threshold corresponding to the maximum cross-region Shannon entropy as the final segmentation threshold. The proposed method is compared with Masi entropy thresholding

Tsallis entropy thresholding

Shannon entropy thresholding

and iterative triclass thresholding on 40 synthetic images and 50 real-world images. The results show that the proposed method is not superior to the 4 compared methods in computational efficiency

but it has significant enhancement in segmentation adaptability and a significant decrease in the mis-segmentation rate.