Color and Gradient Features for Text Segmentation from Video Frames

摘要: Text segmentation in a video is drawing attention of researchers the field image processing, pattern recognition and document analysis because it helps annotating labeling events accurately. We propose novel idea generating an enhanced frame from R, G, B channels input by grouping high low values using Min–Max clustering criteria. also perform sliding window on to group neighboring pixel further enhance frame. Subsequently, we use k-means with k = 2 algorithm separate text non-text regions. The fully connected components will be identified skeleton obtained clustering. Concept component based gradient feature has been adapted for purpose symmetry verification. which satisfy symmetric verification are selected representatives regions they permitted grow cover their respective region containing text. method tested variety frames evaluate performance terms recall, precision f-measure. results show that promising encouraging.