Enhanced Radon Transform Skew Estimation And Correction Algorithm For Scanned Multiple-Choice Forms
Aliyu Muhammad Abdu, Musa Mohd Mokji, Usman Ullah Sheikh, Adamu Ya‘u Iliyasu · The European Proceedings of Social & Behavioural Sciences · 2019
As an extension of our previous work this paper presents an enhanced skew estimation and correction algorithm for multiple-choice (MC) forms. It involves a synergy of the Radon Transform with high precision hit-and-miss algorithm in determining the amount of skew in MC form images, prior to optical mark recognition (OMR) which require properly aligned text lines and edges. As other existing Radon transform based techniques resolve to determining skew angles based on only a single peak regardless of its efficacy, the proposed algorithm first detects the nature of the form (either standard OMR or customized design) and then automatically optimize the selection process for the number of peaks necessary for the estimation of the correct skew. Skew in this context refers to the tilt or in-alignment of edges in degree which is neither parallel nor at right angles to a specified or implied boarder. It is therefore essential to detect and correct the skew at the pre-processing stage in order to avoid perturbation of skew during further processing and extraction of answers. Experiments on various form designs were conducted and an overall accuracy of 98.9% across all experiments has been achieved which shows the superiority of the proposed algorithm. Furthermore, a comparative analysis with other reported algorithms has been made to prove the efficacy of the proposed technique. This technique works well in correcting the skew even with lower resolution images and on those with background variations such as blurring, thus in-line with the future use of phone camera images.