A fake face detection system using a hybrid discrete Chebyshev wavelet  transform and a convolutional neural network.

Authors

The widespread use of artificial intelligence has led to the exploitation of

this development in the development of manipulation processes, for

example, face manipulation with deep learning, which poses a great danger

and threat to information safety and digital security. This work proposes a

new hybrid system in which a new filter extracted from discrete Chebyshev

wavelet transform (DChWT) ( )is effectively integrated with an

artificial convolutional neural network and employed in the process of

detecting face forgeries. The capabilities and characteristics of the

proposed waves were utilized, namely the orthogonality property and the

multi-precision analysis property of waves in the spatial frequency domain

in spatial learning in artificial convolutional networks. The performance of

the proposed system in this work stands out in the superior performance of

the Face Forensics standard, with the achievement of 99.98% accuracy

with a recall value of 99.89%, and 99.91% precision. The confusion

matrix, developed after comprehensive analysis, serves to evaluate the

classification performance of a dataset containing both real and fake

datasets. This evaluation aims to achieve system efficiency by identifying

the hybrid system's strengths in detecting diverse manipulation techniques

and challenges. In this study, the hybrid system outperformed modern

methods by 2.3% in AUC-ROC, achieving high computational efficiency

within the required timeframe and maintaining this efficiency