A Learnable Frequency Gated DCT Denoising Network
DOI:
https://doi.org/10.65455/p24ghq94Keywords:
Image Denoising, Discrete Cosine Transform, Frequency Domain Modeling, Multi Scale Attention, Global Residual, Image RestorationAbstract
DCT-based denoising often relies on fixed rules such as thresholding or retaining only low-frequency components, which easily blur textures or leave residual noise on natural images. To address this limitation, we develop a learnable denoising model that keeps the structure of the DCT domain while allowing the network to adaptively regulate each frequency band. With differentiable DCT/IDCT layers, the method applies trainable masks to high- and low-frequency coefficients, and a lightweight global attention block at reduced resolution provides broader contextual cues at low cost. An image-level residual path further aids reconstruction. Under a compact setup, the model improves PSNR and SSIM from 11.89 dB and 0.2915 (pure DCT) to 25.05 dB and 0.945, showing that replacing fixed heuristics with learnable frequency modulation leads to substantially better restoration quality.References
[1]BUADES A, COLL B, MOREL J M. A non-local algorithm for image denoising//2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. San Diego, CA, USA: IEEE, 2005: 60-65.
[2]AHMED N, NATARAJAN T, RAO K R. Discrete cosine transform. IEEE Transactions on Computers, 2006, 100(1): 90-93.
[3]AHARON M, ELAD M, BRUCKSTEIN A. K-SVD: An algorithm for designing overcomplete dictionaries for sparse representation. IEEE Transactions on Signal Processing, 2006, 54(11): 4311-4322.
[4]ZORAN D, WEISS Y. From learning models of natural image patches to whole image restoration//2011 International Conference on Computer Vision. Barcelona, Spain: IEEE, 2011: 479-486.
[5]GU S, ZHANG L, ZUO W, et al. Weighted nuclear norm minimization with application to image denoising//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Columbus, OH, USA: IEEE, 2014: 2862-2869.
[6]ZHANG K, ZUO W, CHEN Y, et al. Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising. IEEE Transactions on Image Processing, 2017, 26(7): 3142-3155.
[7]ZHANG K, ZUO W, ZHANG L. FFDNet: Toward a fast and flexible solution for CNN-based image denoising. IEEE Transactions on Image Processing, 2018, 27(9): 4608-4622.
[8]ZAMIR S W, ARORA A, KHAN S, et al. Restormer: Efficient transformer for high-resolution image restoration//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans, LA, USA: IEEE, 2022: 5728-5739.
[9] WANG Z, CUN X, BAO J, et al. Uformer: A general u-shaped transformer for image restoration//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans, LA, USA: IEEE, 2022: 17683-17693.
[10]LIU P, ZHANG H, ZHANG K, et al. Multi-level wavelet-CNN for image restoration//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. Salt Lake City, UT, USA: IEEE, 2018: 773-782.
[11]QIN Z, ZHANG P, WU F, et al. Fcanet: Frequency channel attention networks//Proceedings of the IEEE/CVF International Conference on Computer Vision. Montreal, Canada: IEEE, 2021: 783-792.
[12]AHMED N, NATARAJAN T, RAO K R. Discrete cosine transform. IEEE Transactions on Computers, 2006, 100(1): 90-93.
[13]GONZALEZ R C. Digital image processing. New York, USA: Pearson Education, 2009.
