A revised feature extraction method for detecting text page up/down orientation

Based on our previous work, a revised method is proposed for text page up/down orientation detection in this paper. All characters in an scanned image which is from a text page, are isolated by using edge detection algorithm. The feature of each character is extracted through three vertical component runs (VCRs). And the image will be vectorized to a 96-dimensional vector. These samples are trained by using support vector machine (SVM), and a classifier is generated. Experimental results show our proposed method is very effective and its accuracy is considerably higher than some former methods.