Text detection based on improved convolutional neural network and the feature of text lines

In order to solve the problem of low precision in existing natural scene text detection algorithm, a text detection algorithm based on improved convolutional neural network and the feature of text lines was proposed. Firstly, the enhanced maximally stable extreme region (MSER) was used to extract connected components of the image and isolated connected regions were obtained by pruning method. Secondly, an improved convolutional neural network (CNN) was used to eliminate non-character regions and obtain candidate character regions. Thirdly, an algorithm based on the feature of text lines for constructing multi-oriented candidate text lines was proposed to detect arbitrary oriented and curved scene text. Finally, C4.5 decision tree algorithm was applied to classify candidate text lines. Experiments were carried out on ICDAR2013, ICDAR2015 and MSRA-TD500 datasets. Experimental results show that the algorithm can significantly improve the precision and recall of text detection in natural scenes, and is suitable for text in any direction, language and font.