An Optimized Feature Selection Technique in Diversified Natural Scene Text for Classification Using Genetic Algorithm

Natural scene text classification is considered to be a challenging task because of diversified set image contents, presence degradations including noise, low contrast/resolution and the random appearance foreground (font, style, sizes orientations) background properties. Above all, high dimension input image's feature space another major problem in such tasks. This work aimed tackle these problems remove redundant irrelevant features improve generalization properties classifier. In other words, selection qualitative discriminative features, aiming reduce dimensionality that helps achieve successful pattern classification. this work, we use biologically inspired genetic algorithm crossover employed significantly quality multimodal hence accuracy for natural images. The Support Vector Machine (SVM) used average F-Score as fitness function target condition. First after preprocessing images, whole (population) built using representation technique. Second, level fusion approach combine features. Third, F-score classifier, apply meta-heuristic optimization technique GA selection. proposed tested on five publically available datasets results are compared with various state-of-the-art methods. obtained proved performs well while classifying textual non-textual region better than benchmark algorithms.