Akademik Veri Yönetim Sistemi
International Journal of Pattern Recognition and Artificial Intelligence, cilt.39, sa.12, 2025 (SCI-Expanded, Scopus)
Images are corrupted by the Additive White Gaussian Noise (AWGN) from imaging devices. Existing spatial-domain filters often introduce edge distortion at high noise levels. We propose the Inverse Gaussian Harmonic Filter (IGHF), a novel training-free denoising framework that inverts the Gaussian kernel and employs harmonic weighting on standard deviation terms. The first closed-form mathematical model provides superior edge preservation while effectively suppressing noise. We evaluate hybrid approaches combining IGHF with BM3D [K. Dabov, A. Foi, V. Katkovnik and K. Egiazarian, Image denoising by sparse 3-D transform-domain collaborative filtering, IEEE Trans. Image Process. 8 (2007) 2080-2095.] and DnCNN [K. Zhang K, W. Zuo W, Y. Chen, D. Meng and L. Zhang, Beyond a Gaussian denoiser: Residual learning of deep CNN for image denoising, IEEE Trans. Image Process. 7 (2017) 3142-3155.], achieving up to 1.50dB PSNR gain over the state-of-the-art spatial methods. Quantitative evaluation on eight benchmark images demonstrates effectiveness across noise levels (σ2=0.01-0.05). Triple-Hybrid fusion achieves 31.96dB PSNR with superior SSIM (0.912) and edge-preservation FOM (0.850), outperforming BM3D by 0.35dB through systematic grid-search optimization. Runtime analysis confirms real-time capability: Enhanced IGHF processes HD frames in <200 ms with minimal memory footprint (<8 MB), enabling consumer-grade deployment. The deterministic approach provides a novel mathematical foundation for spatial-domain denoising with applications in deep-learning architectures.