Performance Analysis of Optical Character Recognition Using Adaptive Binarization of Degraded Document Images

The planned OCR formula to retrieve the text within the scanned document pictures. Here the text detection formula supported two machine learning classifiers: one permits generating candidate word regions and therefore the different filters out non-text ones. The extract connected elements (CCs) in pictures by victimization the maximally stable extremal region formula. In CC cluster adaboost classifiers are accustomed confirm whether or not the region contains text or not. Then victimization binarization methodology, the grey image is converted into binary image. The binarization outcomes are subject to OCR and therefore the corresponding results evaluated with relevance character and word accuracy. As additional and additional text documents are scanned quick and correct. Extra performance metrics of the proportion rates of broken and uncomprehensible text, false alarms, ground noise, character enlargement and merging. This effectiveness of the planned methodology is additionally confirmed by tests carried on realistic document pictures. For planned formula MATLAB version thirteen package is employed.