Character recognition in low quality document images using local and global features
Chenqiang Gao, Xiaoming Huang · 2014
Although a great success has been achieved for the situation of high quality images during the past decades, Character recognition in low quality images still remains a challenge. To tackle this challenge, in this paper a novel method in the SVM framework is proposed to recognize the characters in low quality document images by using local and global features. Firstly, a multi-scale sliding window strategy with a pruning method of character traits is adopted to generate potential character sub-regions. Then, the conventional global feature and state-of-art local feature, namely histogram of oriented gradients (HOG), are extracted to form the representation of the potential character sub-region. Finally, the Support Vector Machine (SVM) is used to recognize characters with a late fusion strategy. Experimental results show that the proposed method has a better performance even in the situation of existing touched and broken characters situation compared to the conventional method.