ILLUMINATION-INVARIANT OPTICAL CHARACTER EXTRACTION FRAMEWORK FOR REAL-TIME PROCESSING

Abstract

Non-uniform illumination remains one of the most challenging factors affecting the accuracy and reliability of real-time optical character extraction systems. Variations in brightness, shadows, glare, and low-light conditions often lead to poor binarization and incomplete character segmentation. This paper proposes an illumination-invariant optical character extraction framework that enhances and generalizes the adaptive thresholding approach presented in the reference system. The framework integrates illumination normalization, contrast-adaptive preprocessing, locally optimized thresholding, and morphology-based refinement to ensure robust text separation from complex backgrounds. Real-time line, word, and character segmentation modules are designed using optimized projection and region-based methods to maintain stable performance even under dynamic lighting variations. Experimental observations demonstrate consistent extraction quality across diverse illumination conditions, confirming the suitability of the proposed framework for mobile, embedded, and real-time document processing applications. The approach significantly improves accuracy, reduces noise sensitivity, and maintains efficiency for real-time deployments.