Optical Mark Recognition Based on Image Processing Techniques for the Answer Sheets of the Colombian High-Stakes Tests

Optical Mark Recognition (OMR) is the process of electronically extracting intended data from marked fields, such as squares and bubbles fields, on printed forms. OMR is useful for applications in which large numbers of hand-filled forms need to be processed quickly and with a high degree of accuracy, for instance, reading the answer sheets of high-stakes tests. Nowadays, image processing techniques and advancement in computing could help to read the answer sheets, quickly and reducing operational costs. This work introduces a systematic procedure of image processing with two segmentation steps that conclude in the extraction and recognition of marks of answer sheets of Colombian High-Stakes Tests. Some preliminary results show that the accuracy is 99.83%, on average, in the first calibration stage, with 4 blocks of about 400 images, each one. A sampling procedure was performed to determine an adequate number of images to verify the performance of the method in the scenario of the application Saber 11 in the second semester of 2018. The conclusion of the exercise, with around 65.000 images, was 99.7% of accuracy, which was run in an 8 logical processors pc architecture, getting an average speed of 8 sheets per second. Thus making it suitable for real applications or for performing a labeling process for deep learning training.