Image Skew Detection for Formulas without Fraction Bars Using Connected Components Analysis

Skew detection acts as an important role in document image analysis systems. Many methods have been designed for text-dominant document images, but they usually fail in handling formula-dominant images. Hough transform is an effective method to estimate the skew angle of formulas with fraction bars, but cannot dealing with formulas without fraction bars. This paper presents a skew angle detection method for formula images that have not fraction bars, by analyzing connected components. All of connected components in a formula are clustered to two classes based on their widths and heights. The connected components in the class that have good alignment characteristics are selected to calculate the skew angle of the formula. Experimental results on various types of skewed formulas without fraction bars have showed the validity of the proposed method.