Age-Type Identification and Classification of Historical Kannada Handwritten Scripts using Line Segmentation with HOG feature Descriptors

The offline handwritten text recognition is one of the most challenging tasks in document image analysis; our aim is to recreate the cultural importance of the Kannada Language writing tradition through the historical degraded manuscripts. In the present digital era, we need to protect and digitize the resources of our Indian culture and heritage by digitizing those manuscripts which are losing its status; the degraded manuscripts are influenced by weather condition. In this paper, we have attempted to identify and recognise the historical Kannada handwritten scripts of various dynasties; namely, Vijayanagara dynasty (1460 AD), Mysore Wadiyar dynasty (1936 AD), Vijayanagara dynasty (1400 AD) and Hoysala dynasty (1340 AD) by using the improved seam carving text line segmentation method with HOG feature descriptors. The average classification accuracy for different dynasties are computed. The LDA classifier is yielded 93.4%, K-NN classifier has yielded 92% and SVM classifier has 95.5%. Based on the experimentation, the SVM classifier has recorded good classification performance comparatively LDA and K-NN classifiers for historical Kannada handwritten scripts.