Image to Speech Converter – A Case Study on Handwritten Kannada Characters

India is a multi-lingual and multi-content nation containing eighteen authority dialects, Kannada is one among them. A few works have been improved the situation of the recognition of transcribed Kannada characters. The objective of the proposed system is to identify the handwritten Kannada character correctly and convert it to speech to help the children with learning disabilities, like Dyslexia. In this paper, the image features are extracted using contour feature extraction method and k-NN is used for classification. The dataset collected is constrained and has all the basic characters along with compound and complex characters. A total of 5100 basic handwritten Kannada character's (includes Swaras and Vyanjanas) sample images are considered. The total size of dataset comes up to 3 lakh image samples. Pre-processing techniques have been applied to the images. The contour feature extraction method is used for feature extraction. A feature consists of 35 attributes for each image in the dataset. Then these feature vectors are passed to the k-NN classifier for recognition. Each test sample has been recognized as one class based on the nearest neighbors. Once the classes for test samples have been recognized, it is then converted into speech. The proposed algorithm works for all basic Kannada characters. The recognition accuracy is 84.728% considering only moments in the features of test image sample. The recognized character is then converted into speech.