Analysis of an Video for text Detection and Extraction

Abstract

Detection of text characters or text strings in natural scene can deliver significant newlineinformation for various real-time applications. These types of techniques are widely used in real time applications. Various techniques have been proposed for text extraction from natural scene images or videos. These techniques still suffer from the text extraction from natural scene due to newlinediverse pattern of text, camera distortion, background variation, background interfaces and occlusion, illumination, etc. There are numerous issues that need to be taken care to recognize textual data in natural images. Some of the common tasks while performing character recognition are localizing the text, character and word segmentation, character recognition, language model development, etc. In this research we develop a novel visual computing based approach for detection and recognition of texts in natural scene images. The complete process of newlineresearch is split into three module i.e. detection of text utilizing combine feature extraction method, joint feature extraction and recognition of text using deep CNN, and lastly, detection and recognition of text in video scene images. The complete experimental study is carried out newlineusing MATLAB simulation and tested for open source datasets such as ICDAR-2003, ICDAR 2005, ICDAR-2013, ICDAR-2015, Street View Text, IIIT 5k-world dataset, OSTD and MSRA TD 500 dataset. As per the first experiment, a joint feature extraction scheme is presented that considers shape and SIFT feature analysis methods. Here, we present curvature based shape analysis model that is quite different from prevailing standard techniques due to the fact that it does not depend on newlinethe pre-defined detection. Also, to construct the feature descriptor, input text image is passed through canny edge detection process where gradients are computed for each image. Further, local curvature is estimated with the help of gradient and edge map mode. This model is computed for each pixel of input image by computing angular.