Abstract:
Text embedded in videos is of prime importance containing information about the visual
content that can be exploited for efficient indexing and retrieval of videos. Our system
extracts the artificial text and identifies scripts of English and Urdu text scripts appearing in video images. A sequence of unsupervised and supervised techniques is applied. Ground Truth generates an XML file containing information about the text regions as rectangles. In the extraction of text regions, unsupervised approach based on image processing techniques is employed that uses the alignment and geometrical properties of text to get the candidate text regions from the video frames. From the extracted text regions, features vector based on Grey Level Co-occurrence Matrices(GLCM) values is extracted and fed to the Support Vector Machine (SVM) for validation. Script of validated text regions is recognized using texture features based on Local Binary Patterns(LBP). After script identification, XML file is generated and compared with the one generated by the Ground Truth to get the accuracy of the system. System is tested on a number of videos containing English and Urdu scripts of various font sizes and styles. Promising results were generated thus, validating the efficiency of extraction and identification.
