Segmentation of Handwritten University Examination Scripts

The proposed work intends to segment margin and page based on the answers of the handwritten university answer scripts, which further helps in the automatic evaluation of answers based on the respective question number. The documents are segmented based on the contents present inside and outside the margin of the handwritten answer scripts. The dataset was manually collected from various universities. The method of research proceeds with image pre-processing, binarization, and segmentation. The proposed work deals with the bounding box detection which when applied undergoes block segmentation where bounding boxes are formed across each answer block. The system when tested with the respective ground truth images generated an accuracy of 94%. Further the contents inside each bounding box are segmented, extracted, and stored in the directory. As a future work, the segmentedimages obtained can be further used to develop an auto evaluation system.