Hybrid Chinese/English text detection in images and video frames

Abstract

We propose a multiscale texture-based method using local energy analysis for hybrid Chinese/English text detection in images and video frames. Local energy analysis has been shown to work well in text detection, where remarkable local energy variations of pixels correspond to text region or boundary of other objects and lower local energy variations of pixels correspond to background or the interior of non-text objects. Local energy variation is calculated in a local region based on the wavelet transform coefficients of images. Hybrid Chinese/English text in images and video frames can be detected whether it is aligned horizontally or vertically. The font size of text to be detected may vary in a wide range of values. The proposed method has been tested on 321 frame images obtained from local TV programs and a tested dataset with low missed rate and false alarm rate.