News |  Sitemap |  Contact
PDF Export  | 

Douglas H. Ross: Grading exams with ImageJ

Abstract

Often data that needs to be gathered from human respondents can be segmented into multiple responses. Surveys and multiple choice tests are two typical examples. The preferred method is to have the respondents directly enter responses on a computer. If computers are not available responses can be entered on paper and transferred to the computer. Transferring data manually to a computer is tedious, time consuming and error prone. Optical Mark Recognition (OMR) is used to automatically translate marks made on paper forms to computer data. The form is scanned to create an image of the filled in form. The form is registered to assure that the locations of the marks will align with the predefined answer locations. The answer locations are analyzed to determine if a mark was made in that location. Software was created using ImageJ to perform OMR on forms both with and without fiduciary marks.

The form chosen for registration with fiduciary marks was a commercial answer form with answer locations in a grid pattern and fiduciary marks that had a correspondence to answer rows and columns. Using this type of form provided sufficient information that different scanners, different resolutions and some variation in aspect ratio could be accounted for. Initial information about the width and height of the scan was used to scale size and position constants in the software. The steps to register the form were to remove fine detail of the form so that the area of the marks could be evaluated. Six fiduciary marks along the top of the form were used for registration since they had a different area than any other marks on the form. Rotation was by fitting a line through them and translation using the leftmost mark. Having the form registered, the marks along the left side of the form were evaluated for location to determine the row horizontal positions to account for aspect ratio changes. Rectangular areas were defined to correspond to answer locations. Those areas were evaluated to determine if an answer was filled in based on the amount of black.

Registration of the answer form without fiduciary marks was accomplished utilizing the text of the form. The initial operation on this form was opposite to that for fiduciary marks. Large area marks were removed rather than fine detail. Evaluation of rotation was based on the white space between text lines being greatest when the lines are horizontal. A circular area of the form was used. Vertical registration was by projecting all of the text to the vertical axis and then doing a least squares fit with the projection of a standard form. Horizontal registration was by taking a single line of text and projecting it to a horizontal axis and doing a least squares fit with the projection of the same line from the standard form. Answer locations were predefined using ImagJ’s interactive cursor pixel location readout. Answer locations were evaluated to determine the amount of black in a predefined rectangular area.

Keywords

grading, optical, mark, recognition, registration

Administrative data

Presenting author: Douglas H. Ross & Kenneth R. Sloan
Organisation: UAB

co-authors: Douglas H. Ross

Kenneth R. Sloan

Type: Poster (portrait)

© Luxembourg Institute of Science and Technology | Legal Notice