IMPROVED ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM FOR HANDWRITTEN OPTICAL CHARACTER RECOGNITION

Nowadays Handwritten Optical Character Recognition (OCR) has become a lively as well as demanding area of research in the image processing and pattern recognizing departments. The previous system designed an algorithm for training with a hybrid neural network for the OCR which is written by hand. The FLM demonstrated by integrated combination of the two types of algorithm; Firefly algorithm, the other is Levenberg–Marquardt (LM) algorithm in order to train the neural network. At last, the presented the neural network which is derived from FLM is combined among the feed forward neural network, Also, segregation of features is performed depending on the magnitude of information used to train, quantity of hidden neurons and Quantity of hidden layers. However, it only achieves 95% of accuracy. Therefore, there is a necessity to develop a proper character recognition system that must get high precision. In order to resolve this, the proposed system designed an Improved Adaptive Neuro-Fuzzy Inference System (IANFIS) for handling the English hand written character. In this proposed research work, first of all, the median filter is used to remove the noises in the input image, and segmentation is performed. Then the segmented images are processed for obtaining feature sets, positional, and structural descriptors. As the feature sets are obtained, the proposed Improved Adaptive Neuro- Fuzzy Inference System (IANFIS) identifies the handwritten character. Furthermore, to improve the accuracy in recognition Enhanced Step-size based Glowworm Swarm Optimization (ESGSO) algorithm is utilized for learn the parameters. Test results shows that the demonstrated system attains better performance in comparison with the systems that are already in use. The performance is judged with respect to detection rate, Peak Signal-to-Noise Ratio (PSNR) and Signal to Noise Ratio (SNR).