Optimization integrated generative adversarial network for occluded text recognition with language modeling

Abstract

Text recognition has attracted increased attention recently as a result of the complexity of natural settings and the variety of text instances. Various text or character recognition methods are introduced to distinguish the text from the natural scene, but existing methods struggle with the distorted and highly curved text instances. Consequently, an effective method for occluded text or character detection from object-background images is developed using the suggested elephant herding exponential sailfish optimizer-based generative adversarial network (EHESFO-based GAN). In order to build the proposed EHESFO, elephant herding optimization and Exponential SailFish Optimizer (ESFO) are merged. ESFO was created by fusing exponentially weighted moving average and SailFish Optimizer. With GAN, features extracted from the background and foreground of an image are efficiently used for image annotation and text recognition. The best features from the background and foreground images are extracted to create the optimal solution, which increases the efficacy and efficiency of text recognition. While taking the occlusion as 0.4, the proposed EHESFO-based GAN achieved higher accuracy of 98.1090% and lower error of 1.4%.