Improved firefly based convolutional neural network for scene character recognition using bisa algorithm

Abstract

Text, the great invention, has been an important part of human life newlinefrom the dawn of time and it is extremely important in all forms of newlinecommunication. text, on the other hand, is essential for the development of newlinecomputerized real-world applications. as a result, textual data must be newlineextracted from the captured scene images. there are numerous sources newlineavailable, and most applications rely on visual data extraction, such as newlineinformation retrieval, banking, vehicle plate identification, and house number newlinerecognition. moreover, several computerized programs are available to newlineperform scene character recognition. these techniques made the text or newlinecharacter recognition easier from the scene images. usually, the text newlinecharacters in the images are scanned first and converted into a required format newlineto perform further processing of data. nevertheless, the conversion of text in newlinethe images into machine readable form is also challenging when they are in newlinedifferent formats like possessing different orientations, various languages, newlinepresence of both text and numbers, and so on. several recent research studies newlinehave had poor performance because of the presence of noise, and other newlinecircumstances, and the results have degraded due to computational newlinecomplexity. as a result, we proposed two novel techniques for tackling this newlineproblem.in phase-i, a novel approach is known as the improved firefly newlineoptimization-based local trapping-based cnn is utilized to extract features newlinefrom scene images. the exploited iflt can overcome the local trapping newlineissues of cnn. the iflt tunes the hyperparameters of cnn. meanwhile, newlinecnn employs its mlp and alignment layers. the retrieved features are then newlinecategorized using the adoptedsvm based on the appropriate group. the newlineperformance metrics such as average computational time, mean, standard newlinedeviation, and others are analyzed with other existing works to confirm the newlineeffectiveness. thus the proposed work accomplishes a better scene character newlinerecognition approach without any computational complexities. we used six newlinedataset