An Optimized Deep Learning Model For Photographic Images And Photorealistic Computer-Generated Graphics Image Classification

The differentiation between computer-generated images and photographs has emerged as a significant study issue in the image forensics field due to the heightened visual realism of artificial images. Existing research on distinguishing these images typically needs a superior quantity of labeled data for training, high computational resources, and also suffers from inaccurate distinction as well as image quality issues. Therefore, an Adaptive Gazelle Optimization Algorithm Enabled Deep convolutional neural network model (AdGO-DCNN) is developed in this research to resolve the existing difficulties. The DCNN classifier distinguishes the images effectively while reducing the risk of overfitting. The Adaptive Gazelle Optimization (AdGO) helps in the model’s performance improvement with minimized computational time. The experimental outcomes of the research show an excellent performance for the proposed model indicating the model’s efficiency in distinguishing computer-generated and photograph images by attaining an accuracy of 96.54%, 0.09 False negative rate (FNR), 0.07 False positive rate (FPR), 97.04% F-score, 95.18% Precision, and 98.98% recall and 0.964 ROC exceeded other existing approaches.