CONTOURLET TRANSFORM-BASED GENERATIVE ADVERSARIAL NETWORK FOR ENHANCED FACIAL IMAGE DENOISING
Facial image denoising is a critical preprocessing step in computer vision applications such as face recognition, surveillance and biometric authentication, where noise can significantly degrade recognition accuracy and visual quality. While adoption of deep learning techniques particularly the Generative Adversarial Network (GAN)-based approach has significantly improved the image denoising process, integration of GAN with other approaches still enhances it performance especially in preserving fine facial structures such as edges, textures and contours. This study proposes a Contourlet Transform–Based Generative Adversarial Network (CT-GAN) for enhanced facial image denoising by integrating the contourlet transform into a GAN architecture to exploit multi-resolution and multi-directional feature representations for improved noise suppression and structural preservation. The model was evaluated using the FER2013 facial image dataset, where synthetic Gaussian, salt-and-pepper and Poisson noise were introduced at varying intensities to simulate real-world degradation. Experimental results demonstrate that the proposed CT-GAN achieves superior performance compared to conventional CNN-based denoisers and standard GAN models, with improvements in Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM) and Mean Squared Error (MSE). The proposed CT-GAN method achieved a PSNR of 32.1 dB and SSIM of 0.94 alongside the lowest MSE value of 0.005. The integration of contourlet-domain features enhances edge preservation and texture fidelity, resulting in more visually coherent reconstructed facial images.
Authors : Adamu, Z.M., Abdulrazaq, A.A., Sani, B.H., Ismail, A.A., Suleiman, A.A. and Olawoye, T.O.
Category : Open Access Volume (Issue) : 12(1) Date Uploaded : 14th August 2026