TEXT FRAME CLASSIFICATION IN VIDEO

Abstract

A text frame classification scheme is a binary classification scheme which divides a set of images or video frames into text frames and non-text frames. Text frames are the set of images in which text is visible in the image while non-text images do not contain any visible text. Text information in a video or an image or a series of images enables many applications such as event identification in videos, content based indexing and retrieval, traffic sign detection etc. Real time text extraction from video requires that the extraction system does not utilise resources on false alarms in non-text frames. Hence it is important to classify frames into text and non-text frames and only input the text frames to the text extraction system. The objective is to achieve high recall and precision while minimising the computation time. A few existing text frame classification schemes have been surveyed and a novel text frame classification scheme has been proposed. The proposed scheme generates text candidates from the input image by using the Maximally Stable Extremal Regions of the input frame. The non-text components are deleted from the candidate image by using the properties associated with the ellipse equivalent to the component and some other heuristics based rules. The text candidates are then verified for the presence of text by using a piece-wise linearity check on the components. A comparitive study with the existing niethods shows that the proposed method is superior in terms of processing time and offers a comparable classification rate.