The Research And Implementation Of Character Recognition Technology In Videos Based On OCR

ABSTRACT:The most representative information of video content is often the characters in videos. Thus extracting these characters from videos is a significant technology for video retrieval, public opinion monitoring in videos and other applications. Since current commercial OCR software could not recognize text information effectively from video images with complex background and strong interference yet, searching for a universal key frame extraction, text position and character segmentation method to improve recognition efficiency is one of the current research hotspots.Focusing on video key frame extraction technology, this paper presents a key frame extraction algorithm based on corner analysis which is suitable for text extraction Combined with multi-frame merge technology, the algorithm reduces the complexity of background. Compared with other traditional key frame extraction algorithm, it improves the accuracy of key frame with text. Focusing on text region location technology, this paper proposes an algorithm based on combining edge information, intensive corner group information and morphological processing. It uses a double-factor authentication to determine where the text regions are. Compared with other text region location algorithms, the comprehensive performance of this algorithm is better. Focusing on character segmentation technology, this article applies the text-background binarization algorithm based on threshold and improves the original Otsu threshold segmentation algorithm. For each part of the system modules, this paper conducts several comparative experiments of multiple algorithms, selects algorithms suitable for video character recognition or verifies the enhancement of system performance brought by the algorithms and improvements proposed by this article. Finally, this paper also conducts the testing experiment for the performance of the overall video character recognition system.The results of experiments shows that the algorithms and methods presented by this paper are all suitable for video character recognition, each part of the system modules all performs well and the final system reaches a relatively high character recognition rate.