Real -Time Text Detection and Recognition Based on Optical Character Recognition(OCR)

Abstract

Text presented in videos includes vital information for content analysis, indexing, and retrieval of videos. Finding, verifying, and recognizing video text against complex backgrounds is a key technique for extracting this content. The existing system faces challenges ineffectively detecting and recognizing text content in images, limiting its applications to only images and recognizing text in English. This highlights the limitations of the CTPN network, which can only detect text in approximate horizontal directions. Additionally, the CRNN algorithm used for text recognition lacks efficiency in handling occluded text, indicating the need for improvement in both text detection and recognition under complex backgrounds. Our proposed system suggests a method for text detection and recognition in real-time videos, and web cameras with enhanced multilingual support and converts any language text into English. Efficient handling of occluded text, Efficient handling of text orientations, variations of different fontstyles, sizes and text distortions using text detection tools such as the OpenCV. This tool extracts the region of interest of text in the video frames, implementing text detection through Optical Character Recognition (OCR) and Pytesseract. OCR and Pytesseract extract text from video frames and pre-process frames for better recognition, enabling automated text extraction and analysis in videos. This approach offers a promising solution with much more better results than the existing method for detecting text in videos and web cameras.