The flexible and adaptive X-mark detection for the simple answer sheets

Abstract

Template matching is a technique in digital image processing for finding a target object which matches the given template's pattern. This paper mainly aims to apply the template matching technique to check the answer sheet with multiple choices in order to reduce the cost of using an expensive machine called the Optical Mark Reader (OMR) machine as usual. Even more, the machine works with only a high hardness (or dark black) pencils such as ones with the hardness of 2B or higher. The technique of template matching has been applied to detect two diagonal lines of a cross mark (or an X-mark), instead of searching for an X mark which represents a cross mark. For this purpose, two templates representing the two diagonal lines must be tested, with different orientation angles as needed. However, the approach of adaptive matching proposed in this paper also allows a flexible size of the blank slot (a rectangular or square shape in which an X mark may be drawn). The size of a blank slot may be varied between 0.8 to 1.5 centimeters. Even more, the proposed approach also allows some acceptable deviations in terms of the mark shifts from the correct and precise position due to the natural line drawing of a human. Not only does the approach accept a black ink pen or pencil, but also any ink color of pen which is obviously seen comparing with the color of the paper sheet. Thus, these ways make the proposed method more flexible and practical in real use. The total of 6400 slots of the answer sheets are tested with three steps of shifting and rotating. Each step includes 200 marks each of which is acceptable to be a correct X-mark. From the experimental results, the proposed approach provides the accuracy of 100 percent after combining the use of shifted and/or oriented templates while the accuracy of only 42.8 percent is obtained without the combination. However, if a mark is considered as an unacceptable X-mark, the approach can also detect it completely.

A new method of mark detection for software-based optical mark recognition

“…The projection profile and X-mark detection methods were not tested as they can only be used on boxes with ‘X’ marks. In any case, this type of input field is commonly used only in Thailand for public and school examination papers [ 3 , 4 ].…”

“…Recent studies using different implementations of SOMR have reported error rates between 0.02–0.80% [ 3 – 5 , 7 , 17 – 20 ]. The error rate using the new detection method was 0.03% when tested on untrained respondents, which is in the lower end of the range compared to the others.…”

“…The low error rates for the “X-Mark Detection” and “Pixel Counting” methods were achieved only by looking for specific types of marks, which were crosses and completely filled bubbles respectively, while favorable testing conditions were created for TeleForm [ 3 , 5 , 22 ]. The new detection method can achieve similar results under a wide variety of conditions, which these other methods are not able to do.…”

“…Internal labels allow input fields to be closely spaced and appear neater but interfere with the pattern recognition routine required to locate the fields and detect user marks within the fields. In recent literature, there has been much written about new mark detection techniques such as adaptive X-mark matching, projection profiles, contrasting classifiers, finder patterns, and simple thresholding [ 3 – 7 ]. None of these methods however accept internal labels.…”

“…While accuracy is high and detection of colored pen marks is possible, the boxes need to have a minimum size of 0.8 x 0.8 cm. Moreover, rotated marks cannot be consistently detected (57.2% error rate) unless shifted and/or orientated templates are used (0.02% error rate), which greatly adds to the computational complexity of the method [ 3 ].…”