Two Stage Multi Modal Deep Learning Kannada Character Recognition Model Adaptive to Discriminative Patterns of Kannada Characters

Objectives: Designing optical character recognition systems for Kannada is challenging due to higher self-similarity in characters and number of classes. This work addresses the two major problems reduced accuracy false positives characters. Methods: proposes a stage multi modal deep learning technique handle complexity recognition. The are first grouped based on morphological structural similarity. A novel morphological/structural difference maximization convolution kernel trained each group recognize that group. divide conquer strategy reduce model discriminative features Findings: proposed provides 89% which at least 6% compared existing works. solution least10% lower Novelty: sector intensity distribution feature specific curve structure deciding groups Classifier designed Classification done using convolutional neural network with solve similarity problem Keywords: Handwritten Recognition; Deep Learning; Machine Pattern Fuzzy Gaussian