Inclusion of Vertical Bar in the OMR Sheet for Image-Based Robust and Fast OMR Evaluation Technique Using Mobile Phone Camera

Optical Mark Recognition: Advances, Difficulties, and Limitations

“…According to Nguyen et al [28], the translation problem is no longer relevant because with the advent of automatic scanner feeders, the sheets are correctly framed, resulting in a good alignment at the time of digitization. Additionally, cameras have gained relevance to OMR sheet scanning, as observed by us in the proposals by Karunanayake [22], Zampirolli and Gonzalez [46], Rachchh and Gopi [33], in which the solutions are cameras based.…”

“…Reddy et al [35] and Ware et al [42] Pattern recognition and pixel counting Pixels counting in the masked region Dayananda Kumar et al [9] Pattern recognition and imaging filters Using the average intensity and optimal threshold selection Rachchh and Gopi [33] PCA-based skew corrections and pixel counting Counting the binaries pixels and performing one calc with 95% from the minimum Rasiq et al [34] Pattern recognition and image filtering Black pixels counting above 45% Afifi and Hussain [1] Speeded Zampirolli and Gonzalez [46] Eight types totaling 674 tests Shivanna et al [37] A total of 5000 sheets AL-Marakeby [2] 200 images with 30 questions and 4 choices (24,000 choices) Karunanayake [22] Three sets containing 40, 20, 50 answers, with 10 answers each Talib et al [40] Only five examples of students are mentioned in the paper Sanguansat [36] Questionnaire marked by 35 students Patel et al [31] Tested with 310 images Chai [8] Amount of 1000 in 10 sets of 100 sheets. Each answer sheet contains 50 questions and 5 options per question Catalan [7] 800 sheets of 110 choices each Tjahyadi et al [41] Three sets of 27, 1820 and 14 sheets Dayananda Kumar et al [9] Only inform that five different templates with varying number of questions, multiple answer choices, dimension of marking circles, and patterns are processed Rachchh and Gopi [33] 100 images for demo and 40 of real test, all images have 40 choices per sheet Rasiq et al [34] only 5 images are used. Afifi and Hussain [1] 735 sheets and 6 models exams made.…”

“…Rachchh and Gopi [33] and Afifi and Hussain [1] provide the images used in their experiments. Both datasets were last accessed in June 2021, respectively by the following links http:// phc.…”

“…Table 4 shows the average time per sheet or OMR document of 11 solutions in the literature. The overall average processing time not considering Catalan [7] is 2.78 s. However, these numbers should be analyzed carefully since the authors Chai [8] 100.00 0.00 Rachchh and Gopi [33] 100.00 0.00 Loke et al [25] 99.94 0.06 Accuracy reported by combining samples with and without noise AL-Marakeby [2] 99.65 0.35 Reported as average or when reported for an interval, the average value was entered Shivanna et al [37] 99.20 0.80 Afifi and Hussain [1] 98.83 1.17 Average value of two metrics Catalan [7] 98.75 1.25 Zampirolli and Gonzalez [46] 98.70 1.30 Disregards invalid responses in 6.7% of sheets Khan et al [23] 98.00 2.00 Reported as average or when reported for an interval, the average value was entered Dayananda Kumar et al [9] >98.00 <2.00…”

“…This section presents the main restrictions, average processing time, and average processing accuracy found by analyzing the recent literature on software-based OMR. [36] 0.74 Rachchh and Gopi [33] 1.15 Chai [8] 1.40 Karunanayake [22] 2.00 Dayananda Kumar et al [9] 2.50 Patel et al [31] 3.00 Shivanna et al [37] 6.00 Zampirolli and Gonzalez [46] 10.00 Catalan [7] 68.00 Average time 2.78…”