No abstract

As per the studies during many decades, substantial research efforts have been devoting towards character recognition. This task is not so easy as it it appears; in fact humans' have error rate about more than 6%, while reading the handwritten characters and recognizing. To solve this problem an effort has been made by applying the multiple features for recognizing kuzushiji character, without any knowledge of the font family presented. At the outset a pre-processing step that includes image binarization, noise removal and enhancement was applied. Second step was segmenting the page-sample by applying contour technique along with convex hull method to detect individual character. Third step was feature extraction which included zonal features (ZF), structural features (SF) and invariant moments (IM). These feature vectors were passed for training and testing to the various machine learning and deep learning models to classify and recognize the given character image sample. The accuracy achieved was about 85-90% on the data-set which consisted of huge data samples round about 3929 classes followed by 392990 samples.

No abstract