REFLECTANCE-ORIENTED PROBABILISTIC EQUALIZATION FOR IMAGE ENHANCEMENT

Despite recent strides in image enhancement, effectively adjusting brightness and contrast in both low-light and normal-light images remains a challenge for existing methods. Addressing this issue, we introduce an innovative 2D histogram equalization approach. This method posits a dependency between intensity occurrence and co-occurrence, establishing the distribution of intensity occurrence (1D histogram) through the marginalization of intensity co-occurrence (2D histogram). 

By incorporating local pixel value differences in image reflectance into density estimation, our scheme defines the 2D histogram, mitigating the adverse effects of dark lighting conditions and enhancing global contrast more efficiently while minimizing noise amplification. To validate our approach, we conducted evaluations using a dataset, illustrating the superior performance compared to existing studies. Our method demonstrates its ability to significantly enhance the brightness of low-light images without causing over-enhancement in normal-light conditions. 

This innovative 2D histogram equalization technique represents a promising advancement in adaptive image enhancement, particularly in scenarios with varying lighting conditions.

Keywords: 2D histogram equalization, reflectance, Retinex model, contrast enhancement, image enhancement