Enhanced Approach For Handwritten Text Recognition Using Neural Network

Abstract

The off-line character recognition is a process which is used to recognition the pattern. The problem for recognition is to segment the character into isolated word. In this paper, we use a new method which calculates the approximation width of the word for the segmentation. After using this algorithm characters are segmented which may have some extra stokes, braked image. To remove these errors some pre-processing steps are performed to make the character smoother. A new feature extraction algorithm is applied to this segmented character, in this algorithm we used to calculate the pixels range and also exact value of a image. The range value is used for the classification. The classifier has been used to train the neural network. Recognition of a character image we used to calculate the approximation value of this image and the value of this image has tested on the trained neural networks, which shows the recognized character. This system is developed by using JAVA. This system converts the handwritten document into structural text form.