Text detection through hidden Markov random field and EM-algorithm

Abstract

The text is a dominant source and delivers semantic information about a particular content of the respective image or video. Human often gives importance to the text than any other objects in an image or a video frame. Text detection is one of the prime part of the text information extraction process. Text detection process is an exciting and emerging research area in the zone of pattern recognition, and computer vision due to the complex background, illumination, and arbitrary orientation. In this paper, the Hidden Markov Random Field (HMRF) method and Expectation-Maximization (EM) algorithm are employed to detect the arbitrarily oriented multilingual text present in an image or a video frame. The proposed method calculates the max-min cluster to maximize the discrimination between textual and non-textual region. HMRF separates the textual region. EM algorithm maximizes the likelihood of the parameters. Laplacian of Gaussian process is used to identify the potential text information. The double line structure concept is employed to extract the true text region. The proposed method is evaluated on Hua’s dataset, arbitrarily oriented dataset, and horizontal dataset with performance measures recall, precision, and f-measure. The outcome shows that the approach is promising and encouraging.