A CNN based Discrimination between Natural and Computer Generated Images

Abstract

The objective of this research is to enhance classification accuracy by employing a systematic pre- and post-processing pipeline, which includes a Convolutional Neural Network (CNN) approach for distinguishing between computer-generated and natural images. In order to improve the consistency of CNN inputs, the methodology implements preprocessing procedures, including image scaling and normalization. The initial phase entails the extraction of a variety of patterns by utilizing feature selection within CNN layers. Transfer learning employs pre-trained CNN architectures, specifically ResNet, to effectively represent features. ResNet is the optimal choice for classification tasks due to its ability to comprehend complex visual characteristics. In order to improve classification decisions and minimize false positives, a post-processing phase implements thresholding on prediction scores. The experimental results suggest that the proposed strategy is effective and can be implemented in situations that necessitate a high level of precision in distinguishing between authentic and synthetic visual content.