LBP histogram for scale independent Bangla character recognition

Bangla is one of the most used languages around the world. In this paper, we proposed a novel method for classification of printed Bangla characters using a histogram of Local Binary Pattern (LBP). We identified that many existing methods do not handle the character font size properly. We targeted this scaling issue and approach towards it in a new way to recognize Bangla character. Effective region of the character was identified first. Then Local Binary Pattern (LBP) is utilized to get necessary texture information. LBP image is split into various parts and texture information is taken out as a sort of histogram statistic. The dimension of the local histogram is reduced to get an abstract feature from the character using different bins compression technique. All the histograms of different parts are concatenated to make the feature vector. To achieve scale independence, we normalized the histogram (feature vector) to make the shape of the histogram unique for a particular character irrespective of its size. Support Vector Machine (SVM) is used for classification purposes and we achieved a reasonable system accuracy for practical use.