Deep Neural Networks Performance Comparison for Handwritten Text Recognition

Optical Character Recognition (OCR) systems are computer programmers that read text from scanned documents and images. Character recognition and text detection are two texts are analyzed. Characters are classified according to their pattern descriptions or features in the categorization process. The characters are identified using a particular classifier. The unified character descriptor (UCD) to be proposed for characters of the attributes in this environment. After, the matching is used to check that the classification was correct. The appreciation strategy functions well know similar scanned documents, but it cannot distinguish characters with a lot of font distortion and variation. Classifiers based on deep neural networks (DNN) could be used to enhance recognition. The MLP (multilayer perceptron) ensures a high level of recognition when providing thorough training; precision is essential. In addition, the convolutional neural network (CNN) is increasing in popularity because of its strong performance; it has gained. Furthermore, MLP and CNN may both be affected by the training process. We make a comparison amongst MLPs in this study as well as CNN. MLP receives the UCD description as well as the necessary network configuration. We worked for CNN to use a convolutional network developed to recognize machine-printed characters and handwritten (Lenet-5). We modify it to accommodate 62 different classes, including characters and digits. Furthermore, graphic processing unit (GPU) parallelization is examined to speed up both CNN and MLP classifiers on our experiments; we show that while classifying characters, the employed real-time MLP is 2xless relevant than CNN.