Skew Estimation In Document Images Using Wigner-Ville Time-Frequency Functions

In this paper a robust technique for skew detection and estimation based on the application of Cohen’s class distribution using Wigner-Ville time-frequency function, to the projection profile histogram of document images has been proposed. Major strength of the paper lies in choice of the criterion function for skew detection.Instead of considering the average/maximum intensity of Wigner-Ville distribution we use the total intensity to enable detection of very minor changes in the skew angle. Complexity has been substantially reduced by using lesser number of image rotations and considering only a small part of the document. Also rough skew is estimated using morphological means, which further reduces complexity to a large extent. The proposed technique provides good accuracy with a resolution of 0.1 degrees, which is less than the previously reported methods. A wide range of skews varying from +89 to -89 degrees and scripts that include 1 and 2-column pages, text-book pages, handwritten documents, scanned images, newspapers and magazines have been considered for testing. The existing methods, although fairly accurate, have limited range of skews and scripts. It is also found to be less computationally intensive compared to other time -frequency transform based methods.