Text Detection in Multi-feature Fusion Natural Scenes Based on Convolution Deep Belief Network
Yuan Zhong, Jian‐an Fang · Academic Journal of Computing & Information Science · 2020
For the traditional MSER algorithm in the background of complex scenes of text detection will appear in the case of false detection. In this paper, a method of multi-feature fusion of natural scene image pseudo-character filtering based on convolution depth confidence network is proposed. The candidate character text regions obtained from MSER algorithm were extracted and fused with LBP feature, HOG feature and CDBN feature, and finally the characters and pseudo-characters were classified by SVM classifier. And merge the resulting characters to produce the final line of text. Experimental results show that this algorithm can filter out more false character areas and improve the accuracy of text location.