ABSTRACT
Optical Character Recognition (OCR) plays a vital role in converting text from scanned images, covering both handwritten and printed documents. The challenges faced in OCR for Telugu handwritten text are compound characters, a vast character set, limited datasets, character resemblances, and complexities in segmenting overlapping characters. To address the intricacies of segmentation, an algorithm has been devised, prioritizing the preservation of crucial features during character segmentation.For feature extraction we have used ZFNet neural network and demonstrated notable progress in recognition rate. With an impressive character recognition rate of 93.5% and a word recognition rate of 80%, the ZFNet model exhibits robust performance in overcoming the unique challenges of Telugu OCR.
