Scene Categorization From Indoor-Outdoor Images Using Hybrid MAMF-Based Deep Convolutional Neural Networks

Abstract

Image scene categorization is the dominant research area, where the localization of the objects along with the background is performed. At the current scenario, existing classifiers fail to provide the accuracy for the classification. Therefore, a novel approach for image scene categorization is performed using the hybrid features and the Hybrid technique named Mayfly Moth Flame (MAMF) optimization algorithm dependent Deep Convolutional Neural Network (MAMF-based Deep CNN) classifier, which positively impacts on the classification accuracy. This algorithm tunes the classifier towards acquiring the optimal classification performance from the classifier and is developed through interbreeding the characteristic features of the vermins and the caddisflies. The significance of the hybrid features for the classification is implemented and analyzed using the MAMF-based deep CNN classifier. The experimental analysis reveals that the proposed Hybrid features with MAMF-based Deep CNN classifier attains highest accuracy of 96.7215 % and 94.8684 % using SCID2 and SUN-397 datasets, respectively.