THE PERFORMANCE ANALYSIS OF TEMPLATE MATCHING SYSTEM IN LETTER IMAGE RECOGNITION USING ZONING FEATURE EXTRACTION AND INTEGRAL PROJECTION FEATURE EXTRACTION

Dwiyanto, Teofanus and Rusydi, Ilhamdi (2011) THE PERFORMANCE ANALYSIS OF TEMPLATE MATCHING SYSTEM IN LETTER IMAGE RECOGNITION USING ZONING FEATURE EXTRACTION AND INTEGRAL PROJECTION FEATURE EXTRACTION. Other thesis, FAKULTAS TEKNIK.

Preview

Text THE_PERFORMANCE_ANALYSIS_OF_TEMPLATE_MATCHING_SYSTEM_IN_LETTER_IMAGE_RECOGNITION.pdf Download (181kB) | Preview

Abstract

Abstrack Pattern recognition system has been widely applied in pattern recognition, especially the letters image. In letter pattern recognition, feature extraction process is also required to obtain characteristics and specific feature of each image to be recognizable. There are various kinds of extraction of characteristics that can be used in the process of pattern recognition. In this research used two feature extraction of zoning and integral projection. In this study there are several stages of the process undertaken the design and implementation of systems for image object recognition alphabet capital letters A, I, U, E, O, B, C, D, F, G with a font style Arial created with Microsoft Word and printed on paper. The first stage is capturing which is the process of taking pictures the image of the letter. The second stage is the conversion of RGB image of the letter to the image intensity and image intensity will be segmented using bi-level luminance thresholding method. The next process is the labeling and filtering, followed by the process of feature extraction using zoning and integral projection methods and the results of this extraction will be recognized by the template system matching. Tests conducted on 450 images obtained letters from the image of alphabet capital letters A, I, U, E, O, B, C, D, F, G with font size 30, 40, 45, 50, 60, 65, 70, 80 , 85, 90, 100, 105, 110, 115, 120 and the distance between camera and letter image are 15 cm, 20 cm and 25 cm. In this study would also analyze the performance of template matching system for each feature extraction of test data that has previously been used as a database system and the test data outside the database system. Test results show that the template matching will work optimally if the data are used as test data, have previously been used as a database system and the system will work less than optimal if the data are used as test data outside the database system. The percentage of the template matching recognition for testing with test data that has become a database system for each feature extraction is 100% and the average percentage of template matching recognition for testing with test data outside the database system is 67.3% for couples zoning and template matching and 72.2% for couples integral projections and template matching.

Actions (login required)

View Item