Character Skew Estimation: A New and Simple Edge based Model

Abstract

Estimation of character skew is still elusive goal for document analysis group since it has pixels of different directions. It is too essential as it avoids segmentation problems and which in turn reduces computational burden in segmenting text lines and processing whole text line/document to estimate skew angle of the document [18]. In addition, it is useful in correcting arc shaped, circle and zigzag shaped text lines appeared extensively in map and advertisement documents. This paper presents a simple and new model for estimating skew angle for a character based on selection of dominant edges detected by Canny and Sobel edge detector. The proposed model is simple as it works based on just structural information such as compactness of the edges of a character. The model extracts the coordinates of selected dominant edges and it is fed to axis of inertia to compute the skew angle for the character. We have conducted varieties of experiments on synthetic character images and real images with different types. It is revealed that the model works up to + 20deg with more than 3deg deviations. As a result, it raises many research issues in estimating skew for a character. Finally, the results are compared with the results of well known methods to show that proposed model is competitive for text images.