A Deep Convolution Neural Network Based Model for Enhancing Text Video Frames for Detection

Abstract

The main causes of getting poor results in video text detection is low quality of frames and which is affected by different factors like de-blurring, complex background, illumination etc. are few of the challenges encountered in image enhancement. This paper proposes a technique for enhancing image quality for better human perception along with text detection for video frames. An approach based on set of smart and effective CNN denoisers are designed and trained to denoise an image by adopting variable splitting technique, the robust denoisers are plugged into model based optimization methods with HQS framework to handle image deblurring and super resolution problems. Further, for detecting text from denoised frames, we have used state-of-art methods such as MSER (Maximally Extremal Regions) and SWT (Stroke Width Transform) and experiments are done on our database, ICDAR and YVT database to demonstrate our proposed work in terms of precision, recall and F-measure.