Although Optical character acknowledgment (OCR) innovation has accomplished colossal advancement as of late, character misrecognition is inescapable. Keeping in mind the end goal to acknowledge high loyalty substance of archive digitalization, we propose another Convolutional neural systems (CNN) based certainty estimation technique. We distinguish the misrecognized characters through contrasting the certainty esteem and a preset edge, in order to leave the acknowledgment blunders as installed pictures in the yield advanced archives. We received sofmax as the estimation of posteriori likelihood, cover pooling and maxout with dropout innovations in CNN engineering outline. Exploratory outcomes demonstrate that our technique has accomplished an unequivocal change contrasted with pattern framework.