Title: Skew decision process based on machine learning content analysis and clustering
Authors: Tanzila Saba
Addresses: School of Computing, Universiti Teknologi Malaysia, Skudai, Johor, Malaysia; AIDA Lab, CCIS Prince Sultan University, Riyadh 11586, Saudi Arabia
Abstract: This paper presents a robust and fast skew detection approach based on content analysis and projection profile clustering. The proposed approach is quite generic and therefore, could detect skew angle for various types of documents such as graphics, charts, postal labels, handwritten text, forms, drawings and their possible combination. The proposed approach is robust that could deal up to ±89 degree skew angle even scanned at least resolution of 50 dots per inch. Additionally, it is language independent. The proposed technique consists of mainly two steps. In the first step, corners are located except top corner. The skew angle is estimated using the cross correlation of located corners with minimum computational complexity. However, to verify the alignment, horizontal projection profile is analysed. The proposed approach is examined using open access benchmark database of scanned document images. Success rate approaching 100% within a confidence range of 0.3 degrees is reported.
Keywords: skew decision process; automatic information processing; document analysis; semantic web.
DOI: 10.1504/IJCAT.2020.109338
International Journal of Computer Applications in Technology, 2020 Vol.63 No.3, pp.191 - 199
Accepted: 20 Jan 2020
Published online: 03 Sep 2020 *