Fake-Face Image Classification using Improved Quantum-Inspired Evolutionary-based Feature Selection Method

Deep learning models have been quite successful in discriminating synthesized or edited fake-face images. However, in the case of small training data, transfer-learning is rather preferable. This is a complex process for high dimensional feature space due to the curse of dimensionality. To mitigate the same, this paper proposes a new feature selection method for the classification of manually created fake-face images. In the proposed method, a pre-trained deep learning model is used to extract features of an image. Next, an optimal feature subset is selected from the extracted features through an improved quantum-inspired evolutionary algorithm. Lastly, the elicited features are considered to perform the classification. Experiments are conducted on a publicly available manually created fake-face image dataset, namely Real and Fake Face Detection by Yonsei University. The performance of the proposed method is compared with two methods in terms of classification accuracy and the number of selected features. The experimental comparisons exhibit that the proposed method achieves promising results among the considered methods.