Document image binarization under non-uniform illumination remains difficult because shadows, stains, paper texture, and spatial intensity variation weaken the separation between the textual foreground and background. This study presents a controlled comparative and ablation evaluation of a fixed-parameter configuration in which contrast-limited adaptive histogram equalization (CLAHE) precedes Local Otsu thresholding. The configuration was compared with Global Otsu, CLAHE-enhanced Global Otsu, Local Otsu, Niblack, and Sauvola using ten degraded document images and corresponding ground truth masks. Performance was measured using intersection over union (IoU), Dice score, accuracy, precision, recall, specificity, balanced accuracy, and processing time. Sauvola achieved the strongest overall result, with mean IoU and Dice values of 0.4845 and 0.6443, respectively. CLAHE-enhanced Local Otsu obtained mean IoU and Dice values of 0.2941 and 0.4481, respectively, together with high recall (0.9600) but low precision (0.2988). Friedman testing confirmed significant differences among the six methods for IoU and Dice (χ²(5) = 31.943, p<0.001, Kendall's W = 0.639). Holm-corrected Wilcoxon comparisons showed that Sauvola significantly outperformed every other method (adjusted p = 0.0293), whereas CLAHE-enhanced Local Otsu did not differ significantly from plain Local Otsu (adjusted p = 0.4199 for IoU and 0.4219 for Dice). The findings therefore indicate that CLAHE recovered weak foreground strokes but also amplified background degradation and did not provide a statistically validated overlap improvement. Because the experiment used only ten images and no independent validation subset, the conclusions remain preliminary and dataset-specific.................... Keywords: ...............document image binarization; non-uniform illumination; CLAHE; Local Otsu; adaptive thresholding; ground-truth evaluation.