An efficient indoor scene character recognition using Bayesian interactive search algorithm-based adaboost-CNN classifier

Abstract

The primary role in many computer vision applications is text or character recognition in scenes. Under generic conditions, scene text recognition is the most complicated and open research challenge, and numerous scene techniques have been implemented to address this problem. Existing methods encountered a number of challenges during scene character recognition, including complex backgrounds, noise, blur, non-uniform lighting, local distortion, and different fonts. Hence, we present Bayesian interactive search algorithm (BISA) with AdaBoost-based convolutional neural network (BISA with AdaBoost-CNN) for scene character recognition to tackle the former issues. The word to consecutive conversion and scene character recognition are the two key components in the proposed work. At first, the HOG and SIFT feature descriptors are extracted in word to consecutive conversion. Next, the Bayesian interactive search algorithm (BISA) is utilized to enhance the presentation of AdaBoost-based convolutional neural network (BISA with AdaBoost-CNN) for scene character recognition. Experimentally, different kinds of evaluation measures are used thereby the implementation works handled in MATLAB software. The proposed BISA with AdaBoost-CNN outperforms higher recognition accuracy than other existing approaches.