Mobile Captured Document Classification based on Blind Image Quality Assessment Model using Brisque

Assessment of document image quality helps in applying suitable pre-processing methods for enhancing document images. The result of enhancement leads to improved recognition rates by OCR. This paper proposes a mobile-captured document classification model based on the blind image quality score. The quality scores are generated by fitting the document wise opinion scores with the Brisque metric. Document wise opinion scores are collected using subjective method. Initially, learning models are created based on the document wise opinion scores. Two methods are proposed as part of this study to generate the document opinion scores, one is using a subjective assessment method, and the other through a random value generation function. The attempt to fit the created learning models using the most dominantly used Brisque's blind image quality measure. The classification of mobile-captured handwritten documents is performed solely using of blind quality score generated by Brisque using simple linear classifiers. Datasets of mobile-captured handwritten documents of shadow, blur, and good readability are considered for experimentation. The evaluation of proposed models for training and test sets shows that the learning models created with subjectively assessed opinion scores produced improved results.