Text Extraction from Video Images

摘要

Video data contains beneficial textual information such as scene text and caption text. The different types of videos like movies, news videos, and TV programs video etc. are created by various video frames based on its purpose. In a country like India, there are only fewer studies has done on text extraction from video data especially in south Indian languages like Malayalam, Telugu, Kannada, and Tamil. The extracted text has many useful applications in video indexing, video key searching and assisting visually challenged people. Malayalam news channel named 'Mathrubhumi News' videos data are considered for the proposed study. It is very beneficial to Kerala people as it is one of the most media-centric regions in the world. In this proposed paper, a new method for text extraction experiments. The anticipated method extracts 13 different features for classifying the image consists of text or not. Both spatial and frequency domain features are extracted to classify. The different types of classification techniques are used to validate the algorithm. Simple Logistic, J48 and Random Forest classification techniques are giving a good result when compared to other methods. Results are encouraging, the average success rate found to be 98%.