Adaptive 2D Entropy-Based Image Segmentation for Maximum Gray Levels in the Top-Left 5 Pixels of 3×3 Neighborhoods

This study introduces a 2D entropy framework for adaptive image segmentation through discriminative 2D histogram profile generation. Conventional segmentation systems predominantly utilize central pixel intensity values and spatial averaging techniques within static neighborhood windows, which inherently fail to account for directionally oriented texture characteristics. Existing entropy-based metrics, constrained by oversimplified linear sampling along horizontal or vertical axes, exhibit limitations in modeling holistic inter-pixel dependencies due to their restricted spatial sampling schemes. To overcome these constraints, we propose a dual-feature descriptor that integrates two spatially adaptive components: (1) the central pixel intensity and (2) the maximum gray-level value within a strategically selected top-left 5 pixels of its 3×3 neighborhood. This hybrid feature extraction mechanism enhances local grayscale distribution characterization while maintaining computational efficiency. Empirical validation through comprehensive image experiments using the MATLAB Image Processing Toolbox demonstrates that our proposed method achieves performance comparable to traditional approaches.