Segmentation-Free Character Recognition System using Direct Cosine Transform Features

Segmentation-Free Character Recognition adds to Optical Character Recognition by precluding the need for separating the individual characters, a step which tends to involve errors in difficult scripts like Malayalam. Optical Character Recognition processes are severely constrained by uneven character spacing, overlap structures, and contextual dependencies that decrease recognition precision. Current methodologies like template matching, neural networks, and common Hidden Markov Models are ineffective in dealing with these complexities to a satisfactory degree. This work presents a method that uses Discrete Cosine Transform for character sequence modeling and noise elimination, and Hidden Markov Models for feature extraction and noise elimination. The suggested method greatly enhances the stability and accuracy of character recognition, especially for handwritten Malayalam text. Experimental results prove the efficiency of this method in real-world usage. The approach is scalable and applicable for document digitization, transcription, and archival applications, adding to the general field of automated text recognition in complicated script-based languages.