An Intelligent Handwritten Malayalam Text Recognition Using Optimized Machine Learning Based Optimization Frameworks

Traditional Machine Learning (ML) and Deep Learning (DL) models are failed to predict the characters, overlapping sentences and diacritical marks. Because Malayalam has various basic compound characters which are more than 50. Also, generalization is failed with handwriting variations for different individuals. In some cases, frequent misclassification is occurred due to the similar characters. Moreover, larger training dataset has required higher computing power so, the efficiency was degrading. Therefore, to develop an intelligent framework which is used to extracts the Malayalam text from the images. To improves the pre-processing stage by adding the entire steps such as normalization, binarization, augmentation and noise removal. To overcome the Malayalam script complexity such as visually same characters, diacritical marks, etc. This paper develops a novel Bayesian Neural Network based Hybrid Cuckoo Search with Lyrebird (BNN-HCSL) algorithm to enhance the character recognition efficiency. Moreover, the developed BNN-HCSL replica was implemented on MATLAB platform and the comparative analysis was performed in terms of performance matrices such as accuracy, precision, recall, F-Measure and Word Error rate (WER) to analyse the effectiveness of the prediction. Moreover, the attained outcomes are demonstrates the better for proposed model in attaining accurate and scalable Malayalam handwritten image to text recognition. Also, this study has contributes the document digitization by accessing the Malayalam handwritten contents.