13019

11897

كارشناسي ارشد

مخابرات سيستم

اصفهان: دانشگاه صنعتي اصفهان، دانشكده برق و كامپيوتر

۱۳۹۶

سيزده،۷۳ص.: مصور، جدول، نمودار

رسول اميرفتاحي

محمدتقي صادقي

بهزاد نظري، محمدرضا احمدزاده

1396/09/01

كتابنامه

برق و كامپيوتر

مهندسي برق و كامپيوتر

ID11897

Separation and Recognition of English Handwritten Digits in noisy conditions Sanaz Nasiri Dahaj Date of Submission Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language FarsiSupervisor Dr Rasoul Amirfattahi Advisor Dr Mohammad Taghi SadeghiAbstractToday handwritten digits recognition plays a pivotal role in industrial applications Despite lots ofresearches have been conducted in this field it is still a state of the art research line In general ahandwritten digit recognition system consists of binarization discrimination feature extraction andclassification blocks First step to digit recognition is to subtract image background and to keep pix els related to digits Different algorithms have been proposed so far which can be categorized intocluster based and threshold based methods Recognition systems require various segments of the im age which can be acquired by either external segmentation or internal segmentation For dimensionreduction in feature extraction step and description of data in feature space different statistical struc tural or transform domain methods have been utilized The last step of handwritten digit recognitionis classification of handwritten digits Popular algorithms of K Nearest Neighbor KNN NeuralNetwork NN and Support Vector Machine SVM have been used for classification and test datalabel recognition In recent years sparse representation based approaches for classification havealso been proposed In this thesis various algorithms of handwritten digit recognition is analyzedin order to propose a suitable method for automatic handwritten digit recognition of slabs in IsfahanSteel Co These images have noisy non uniform background and digits are destructed by saggingof the color Therefore by analyzing various binarization methods a K means cluster based algo rithm which has higher accuracy than Otsu Niblack Sauvola and multi scale grid based Sauvolawas utilized in our method In addition by using preprocessing methods shadow removal saggingelimination and post processing the quality of digits and their complex background improves whichleads to more convenient digit segmentation For classification of data a sparse representation basedclassifier is proposed The accuracy of proposed algorithm is 98 95 for MNIST standard databaseand 81 17 for Isfahan Steel Co database Keywords handwritten digit recognition slab binarization segmentation feature vector clas sification

رسول اميرفتاحي

محمدتقي صادقي

بهزاد نظري، محمدرضا احمدزاده