Computer Vision Approach for Detection and Extraction of Text from Video Frames Using Color Continuity, Color Variations and HV Projections Methods

Ramgoapl Segu (Dayanand Sagar College of Engineering, Bengaluru, India); K Suresh (SEA Engineering College, Bengaluru, India)

Multimedia data retrieval and indexing is an important aspect for information processing and information extraction. In the field of multimedia data indexing, video data indexing is considered as a challenging task for various type of real-time applications. In order to improve the retrieval and indexing performance, text detection and extraction has gained attraction from research community. In this work, we presented text detection and extraction process from video scene frames. In order to accomplish the desired objective, the complete process is divided into two stages as (a) text detection and (b) text recognition. For text detection, a combined approach is presented where color continuity, color variations are computed and later text localization is performed using horizontal and vertical projections. In the next phase, text recognition is performed where DWT based features are extracted and trained using RBF classifier. The complete experimental study is carried out using MATLAB simulation and tested for open source datasets. The comparative study shows that the proposed approach achieves better performance when compared with the existing techniques.

Journal: International Journal of Simulation- Systems, Science and Technology- IJSSST V20

Published: Feb 28, 2019

DOI: 10.5013/IJSSST.a.20.01.07