A segmentation approach based on structured learning for recognition preprocessing
Abhijit Joshi, Deeksha Bharadwaj · 2016
Segmentation of manually written record photos into content material traces and words is a crucial undertaking for optical character acknowledgment. However, for the reason that elements of written by hand record are abnormal and various depending at the character, so far it is considered as a challenging issue. A good way to deal with the hassle, we formulate the word segmentation problem as a binary quadratic undertaking problem that considers pair-wise relationships among the crevices and in addition the probabilities of individual holes. No matter the fact that sever a parameters are covered in this detailing, we gauge all parameters taking into account the based Naive Bayes classifier so that the proposed technique features admirably paying little admire to composing styles and composed dialects without patron characterized parameters. Test effects on ICDAR 2009/2013 penmanship division data-set exhibit that proposed method accomplishes the high-quality in magnificence execution on Latin-primarily based and Indian languages.