Multilevel Image Thresholding and Handwritten Chinese Character Recognition by Support Vector Machines
碩士 === 國立屏東商業技術學院 === 資訊工程系(所) === 99 === Recently, support vector machines (SVM) has been adopted in various categorization applications, such as handwritten character recognition, face recognition, and text categorization. The experimental results of these researches confirmed the effectiveness of...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Others |
Language: | zh-TW |
Published: |
2011
|
Online Access: | http://ndltd.ncl.edu.tw/handle/35422744004406021613 |
id |
ndltd-TW-099NPC05392003 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-TW-099NPC053920032015-10-13T20:22:50Z http://ndltd.ncl.edu.tw/handle/35422744004406021613 Multilevel Image Thresholding and Handwritten Chinese Character Recognition by Support Vector Machines 以支撐向量機為基礎之影像多值化處理及手寫中文辨識 Yi-Guei Lin 林義貴 碩士 國立屏東商業技術學院 資訊工程系(所) 99 Recently, support vector machines (SVM) has been adopted in various categorization applications, such as handwritten character recognition, face recognition, and text categorization. The experimental results of these researches confirmed the effectiveness of SVM. The mathematical theory used in SVM is related to the subfield of nonlinear programming, quadratic programming, which has been developed by many researches in the machine learning area to forther improve the performance of SVM. The research first proposed a SVM-based bi-level image threshoding method, which effectively used the features of the selected pixels in the gray-level image, including coordinates, gray level and gradient. The trained SVM can binarize all pixels in the gray-level image to get a resultant binary image. Experimental results reveal that the proposed method can effectively binarize a gray-level image, even an uneven-lighting image. Based on the proposed binarization method, the research proposes a new multilevel image thresholding method, which recursively transform the gray-level image into the resultant image with the desired pixel levels. Experimental results reveal that the proposed method can categorize the pixels with the same gray level adaptively, and the resultant multilevel image is promising. The research also proposes the SVM-based handwritten Chinese character recognition method. It first extracts the features for an input handwritten character. The coarse classification using mean feature vectors then gets the candidate characters. Finally, the SVM is trained by the training handwritings of the candidates and then determines the category of the input handwriting. According to the experiments, the rate of recognizing handwritten Chinese characters can be increased to 98.31%. Cheng-Huang Tung 董呈煌 2011 學位論文 ; thesis 118 zh-TW |
collection |
NDLTD |
language |
zh-TW |
format |
Others
|
sources |
NDLTD |
description |
碩士 === 國立屏東商業技術學院 === 資訊工程系(所) === 99 === Recently, support vector machines (SVM) has been adopted in various categorization applications, such as handwritten character recognition, face recognition, and text categorization. The experimental results of these researches confirmed the effectiveness of SVM. The mathematical theory used in SVM is related to the subfield of nonlinear programming, quadratic programming, which has been developed by many researches in the machine learning area to forther improve the performance of SVM.
The research first proposed a SVM-based bi-level image threshoding method, which effectively used the features of the selected pixels in the gray-level image, including coordinates, gray level and gradient. The trained SVM can binarize all pixels in the gray-level image to get a resultant binary image. Experimental results reveal that the proposed method can effectively binarize a gray-level image, even an uneven-lighting image. Based on the proposed binarization method, the research proposes a new multilevel image thresholding method, which recursively transform the gray-level image into the resultant image with the desired pixel levels. Experimental results reveal that the proposed method can categorize the pixels with the same gray level adaptively, and the resultant multilevel image is promising. The research also proposes the SVM-based handwritten Chinese character recognition method. It first extracts the features for an input handwritten character. The coarse classification using mean feature vectors then gets the candidate characters. Finally, the SVM is trained by the training handwritings of the candidates and then determines the category of the input handwriting. According to the experiments, the rate of recognizing handwritten Chinese characters can be increased to 98.31%.
|
author2 |
Cheng-Huang Tung |
author_facet |
Cheng-Huang Tung Yi-Guei Lin 林義貴 |
author |
Yi-Guei Lin 林義貴 |
spellingShingle |
Yi-Guei Lin 林義貴 Multilevel Image Thresholding and Handwritten Chinese Character Recognition by Support Vector Machines |
author_sort |
Yi-Guei Lin |
title |
Multilevel Image Thresholding and Handwritten Chinese Character Recognition by Support Vector Machines |
title_short |
Multilevel Image Thresholding and Handwritten Chinese Character Recognition by Support Vector Machines |
title_full |
Multilevel Image Thresholding and Handwritten Chinese Character Recognition by Support Vector Machines |
title_fullStr |
Multilevel Image Thresholding and Handwritten Chinese Character Recognition by Support Vector Machines |
title_full_unstemmed |
Multilevel Image Thresholding and Handwritten Chinese Character Recognition by Support Vector Machines |
title_sort |
multilevel image thresholding and handwritten chinese character recognition by support vector machines |
publishDate |
2011 |
url |
http://ndltd.ncl.edu.tw/handle/35422744004406021613 |
work_keys_str_mv |
AT yigueilin multilevelimagethresholdingandhandwrittenchinesecharacterrecognitionbysupportvectormachines AT línyìguì multilevelimagethresholdingandhandwrittenchinesecharacterrecognitionbysupportvectormachines AT yigueilin yǐzhīchēngxiàngliàngjīwèijīchǔzhīyǐngxiàngduōzhíhuàchùlǐjíshǒuxiězhōngwénbiànshí AT línyìguì yǐzhīchēngxiàngliàngjīwèijīchǔzhīyǐngxiàngduōzhíhuàchùlǐjíshǒuxiězhōngwénbiànshí |
_version_ |
1718046043417870336 |