Multiple-level threshold for image partitioning exploiting median greyscale-based 2D entropy in 4-neighbourhood

The application of 2d entropic metrics facilitates adaptive thresholding partitioning through making 2d histogram. Legacy approaches predominantly rely on central intensity determination and arithmetic average calculation within neighbourhood. Empirical evidence however demonstrates that orthogonal directional disparities in tonal distribution inadequately characterize inter-pixel relationships across 4-neighbourhood. To address this limitation, this study proposes a median-greyscale paradigm that constructs two-d histograms using both central intensity values and the median luminance detected within 4-neighbourhoods. Comprehensive evaluations across entire of MATLAB image toolbox datasets validate this novel framework's superiority, achieving obvious improved partition visual effects compared to conventional methods.