<![CDATA[Manual correction of handwritten exam answer sheets is time-consuming, especially when the number of students is large and the questions are essays requiring in-depth answers. This study designs and develops a website-based system called SiPena that integrates three main components: image preprocessing using the OpenCV library with grayscale conversion and Otsu's thresholding, handwriting recognition using the Convolutional Neural Network (CNN)-based EasyOCR library, and automatic grading using the Fuzzy String Matching algorithm with the token_set_ratio method from the FuzzyWuzzy library. The system also provides a full-sheet grading mode that allows lecturers to upload a single photo of an answer sheet containing all questions at once, as well as a manual override feature to correct automatic grading results. Black-box testing of the system's 15 core functions showed that all functions run according to the established success criteria, while white-box testing confirmed that the logic flow of the preprocessing and grading pipeline, including three answer-parsing strategies and a tiered scoring scheme, runs correctly. A Character Error Rate (CER) comparison test using 210 photos of handwritten answer sheets from 29 respondents was conducted under two conditions: EasyOCR without preprocessing (Condition A) and grayscale conversion with Otsu's thresholding before EasyOCR (Condition B). The test results show an average CER of 42.24% for Condition A and 42.57% for Condition B, a difference of only 0.33%, indicating that the preprocessing stage does not significantly reduce OCR accuracy and even improves recognition accuracy for 17 of the 29 respondents. Grading accuracy testing on the same 210 answers showed a system accuracy of 96.67% compared to the lecturer's manual grading, with an average Fuzzy Matching similarity score of 71.94 out of 100, indicating that the system is able to automatically grade handwritten essay answers with a high level of consistency.]]>
