Skew correction and image alignment for accurate region of interest detection in scanned exam papers

Authors : Ali Şentürk

Pages : 1-9

Doi:10.55974/utbd.1637840

View : 74 | Download : 111

Publication Date : 2025-11-27

Article Type : Research Paper

Abstract :Accurate digit segmentation is a critical process in handwritten digit recognition. In structured documents, digits are written in predefined locations based on template files. One common example is exam papers, where students’ identification numbers and evaluation grades are written in designated regions. However, in scanned documents, these locations are often misaligned due to skews, which negatively affects segmentation accuracy. This study proposes a skew detection and correction method combined with template matching based image alignment to improve digit segmentation for handwritten digit recognition. Unlike general-purpose methods, our approach focuses on structured exam templates, ensuring that numeric entries like student IDs and question grades are accurately extracted. Automating this process is particularly valuable for grading since manual entering scores for each question is a labor-intensive task, especially in large classes. Experimental results on 211 exam papers containing 3,407 handwritten digits show that 2,462 (72%) corrections were required due to misalignment. With the proposed alignment method, this number is reduced to only 333 (9.7%), demonstrating its effectiveness in template-based handwritten digit recognition.
Keywords : Görüntü işleme, Çarpıklık düzeltme, Rakam bölütleme, Rakam tanıma

ORIGINAL ARTICLE URL