Edge-aware Image Smoothing with Relative Wavelet Domain Representation
Authors: Huiqing Qi, Xiaoliu Luo, Tingting Li, Fang Li
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments substantiate that our method has a competitive superiority with previous algorithms in edge-preserving and artifact removal. Visual and numerical comparisons further validate the effectiveness and efficiency of our approach in several applications of image smoothing. To assess the performance and effectiveness of our approach, we conduct a comparative analysis against several existing methods... We utilize five no-reference image quality objective evaluation metrics... We also employ two widely recognized quality metrics, PSNR and SSIM, to assess the images resulting from compression artifact removal... Ablation Study. To assess the capability of RWDR in distinguishing between textures and structures, we conduct an ablation study on RWDR in Figure 8. |
| Researcher Affiliation | Academia | Huiqing Qi School of Electronic & Electrical Engineering Nanyang Technological University EMAIL Xiaoliu Luo College of Science Chongqing University of Technology EMAIL Tingting Li School of Mathematics & Information Science Zhengzhou University of Light Industry EMAIL Fang Li School of Mathematical Sciences, Key Laboratory of MEA(Ministry of Education) & Shanghai Key Laboratory of PMMP East China Normal University EMAIL |
| Pseudocode | Yes | Algorithm 1 Mutually Guided Edge-aware RWDR Input: Image I, σe, γ, λe, rmin, rmax, σ, ε, λ, K, ai, σr. Initialization: F 0 = I. 1: Calculate E via Equation 15; 2: Calculate m via Equation 17; 3: Calculate π with m and E via Equation 18; 4: Calculate ωp with π in Equation 19; 5: for t = 1 to T do 6: Update St with F t 1 via Equation 12; 7: Update F t+1 with St and ωp via Equation 13; 8: end for Output: Smoothed image S. |
| Open Source Code | Yes | Our code will be available on my github. |
| Open Datasets | No | The paper refers to using "clip-art images" and natural images for experiments, but does not provide specific names, links, DOIs, or formal citations for these datasets to confirm their public availability. It also mentions using objective evaluation metrics from cited papers, but these are metrics, not datasets. |
| Dataset Splits | No | The paper mentions using a "10% compression rate for the clip-art images" but does not provide any specific information about training, validation, or test dataset splits, percentages, or sample counts. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware used for running the experiments, such as GPU models, CPU types, or memory specifications. |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers, such as programming languages, libraries, or frameworks used for implementation. |
| Experiment Setup | Yes | For the scheme presented in Algorithm 1, the input parameters are σe, T, γ, λe, rmin, rmax, σ, ε, λ, K, ai, and σr that need to be initialized in advance. σe, γ, λe, ε, and K are fixed parameters. We empirically found that smoothed results are not sensitive to these parameters. The default reasonable fixed settings are σe = 3, γ = 10/255, λe = 4, ε = 5e 4, and K = 4, which are suitable for each test image. We have set rmin = 0 to guarantee that the minimized value in the edge-aware scale map is close to 1. However, rmax is changed as the width of textures for different images. The recommended value of rmax lies in the range [5, 15]. We empirically find that different strengths of weak edges can be preserved with different values of σr. The recommened range of σr is [10, 30]. For our experiments, we have used σ = 3, and λ is in the range [0, 0.008]. We utilized a1 and a3 of Equation 8 in our all experiments. ... The parameter T is crucial for the quality of smoothed images... we set the max iteration number T = 4 in all our experiments. |