Text recognition from image using artificial neural network and genetic algorithm

An important problem in text recognition such as handwritten or character images from the text are difficult to read. The decoding of these texts has important applications in many areas. Many approaches have been proposed for solving the text recognition or classification problem. We propose an artificial neural network and genetic algorithm to solve effective text recognition problem. A hetero-associative neural network is proposed to train the system for deciphering digits from pdf or jpeg images which are not readable. Also, a crossover based genetic algorithm is proposed for deciphering texts from handwritten or text file in image form. The main objective is to convert the text data from PDF and deciphering into digits so that characters are recognized easily. The proposed genetic algorithm repeatedly performs crossover on sections and parts of text data from an image file to train the system. The genetic algorithm after training with text data in image converts it into a form that can be recognized easily. The algorithm solves the problem of deciphering digits and characters from image by parsing image and converting it to a pixel array. The algorithm selects digits and characters and performs crossover with trained patterns with variable heights. This provides best possible ways to discover the image pattern from trained patterns. In this paper, a comparison of neural network and GA with crossover has been done. The proposed approach gives the results which are significant encouraging.