Pixel-Based Unsupervised Classification Approach for Information Detection on Optical Markup Recognition Sheet
Enoch Opanin Gyamfi, Yaw Marfo Missah · Advances in Science Technology and Engineering Systems Journal · 2017
This paper proposed an Optical Markup Recognition (OMR) system to be used to detect shaded options of students after MCQ-type examinations.The designed system employed the pixel-based unsupervised classification approach with image pre-processing strategies and compared its efficiencies, in terms of speed and accuracy, with object-based supervised or unsupervised classification OMR systems.Speed and accuracy were tested using asymptotic running time and confusion matrix, respectively.The study began by involving the ideas of 50 sampled students in the design of an OMR template to be used by the proposed system.The study used six accuracy parameters to compute the effects of the three image pre-processing strategies, two-dimensional median filtering, contrast limited adaptive histogram equalisation, scanlines and standard Hough transform techniques.These strategies proved to increase the accuracy rates of the proposed system.The study finally proposed strategies to detect shaded circle bubble with its centre and block neighbouring pixels within it.These labels were stored in row-by-column one-dimensional array matrices.The study then concluded that the proposed pixel-based untrained classification OMR algorithm, is statistically fast and accurate than the object-based untrained classification OMR algorithms.