IMPLEMENTASI RESILIENT PROPAGATION NEURAL NETWORKDALAM APLIKASI OCR DAN OMR UNTUKPEMBACAAN LEMBAR NILAI

• Main Author: Teja Sophista V Rusyana, - (Author)
• Format: Book
• Published: 2011-04-27.
• Subjects: Thesis; NonPeerReviewed
• Online Access: Link Metadata
• Notes: http://repository.upi.edu/105683/3/s_kom_0608494_table_of_content.pdf; http://repository.upi.edu/105683/2/s_kom_0608494_chapter1.pdf; http://repository.upi.edu/105683/6/s_kom_0608494_chapter2.pdf; http://repository.upi.edu/105683/1/s_kom_0608494_chapter3.pdf; http://repository.upi.edu/105683/4/s_kom_0608494_chapter5.pdf; http://repository.upi.edu/105683/5/s_kom_0608494_bibliography.pdf
• Keywords: L Education (General); QA76 Computer software

Students' grade data filing at Universitas Pendidikan Indonesia (UPI) is conducted by Biro Administrasi Akademik dan Kemahasiswaan (BAAK) using Assessment Form that has been filled by the lecturer and returned to BAAK to be processed with Optical Mark Reader machine, the yellow-colored copy of Assessment Form are kept by the lecturer as an archive. All the students' grades data is stored at BAAK, so if the lecturer needs digital data of the college students' grades, the lecturer must request data to BAAK, buy an expensive OMR machine, or entry the data from Assessment Form manually. Therefore, this research produces software that can retrieve data automatically from the yellow-colored copy of Assessment Form with an ordinary image scanner. There are two kind of data retrieval used by this software, Optical Character Recognition (OCR) to read the Lecturers Code, Course Code, the NIM, and Optical Mark Reader (OMR) to read mark of the grades. To measure accuracy level there are four kind of feature extractions are tested in this OCR research; Matrix Matching, Vertical/Horizontal Projection, Fundamental Features, and Hybrid. The output of this feature extraction will be used as input to the Artificial Neural Network and will be trained using Resilient Propagation to adjust its weight. As for OMR, used a simple method of pixel counting to determine where the mark is located. In the first experiment, with the expected error level less than 0.0001 or 0.01%, with Resilient Propagation training, the Matrix Matching method is failed, Vertical/Horizontal Projection required 190 iteration, Fundamental Features required 195 iteration, and the Hybrid required 97 iteration. As for the reading the actual data test, accuracy of Matrix Matching is less than 70%, accuracy of Vertical/Horizontal Projection is 94%, Fundamental Features 94% and Hybrid 96% of accuracy, as for OMR accuracy level reach 100% under suggested circumstances.