Detail preserving image denoising with patch-based structure similarity via sparse representation and SVD

Shi, Miaowen; Zhang, Fan; Wang, Suwei; Zhang, Caiming; Li, Xuemei

doi:10.1016/j.cviu.2021.103173

Cited by 27 publications

(8 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this section, we conduct extensive experiments to demonstrate the performance of the proposed denoising algorithm (TSLR). In addition, we compare our algorithm with some exsiting non-deep denoising algorithms, including BM3D [6], NCSR [5], SAIST [34], WNNM [12], LIIC [15], RM [16], BMLSVDTV [25], DPID [47], and deep learning-based denoising algorithms (e.g., Dn-CNN [19], FFDNet [44] and MLEFGN [45]). Three test datasets are used to evaluate the AWGN variance σ ∈ {10, 15, 25, 30, 50, 100}.…”

Section: Resultsmentioning

confidence: 99%

“…In this subsection, the denoising performance of the proposed TSLR algorithm is first verified and evaluated using the natural images from Set12 dataset. In all experiments, the recovery quality of images is evaluated according to two objective standards: Peak Signal to-Noise Ratio (PSNR) and Structural Similarity index (SSIM) [67], as shown in Tables 2 and 3, together with those from BM3D [6], NCSR [5], SAIST [34], WNNM [12], LIIC [15], RM [16], BMLSVDTV [25], and DPID [47]. For each algorithm, "AVE." represents the average PSNR and SSIM values of all recovered images.…”

Section: Experimental Results Compared With Non-deep Methodsmentioning

confidence: 99%

“…PSNR(dB) comparisons of BM3D [6], NCSR SAIST [34], WNNM [12], LIIC [15], RM [16], BMLSVDTV [25], DPID [47]…”

Section: Tablementioning

confidence: 99%

“…SSIM comparisons of BM3D [6], NCSR [5], SAIST WNNM [12], LIIC [15], RM [16], BMLSVDTV [25], DPID [47], and TSLR (Ours) on Set12 dataset Schemes BM3D NCSR SAIST WNNM LIIC RM BMLSVDTV DPID TSLR (Ours) BM3D NCSR SAIST WNNM LIIC RM BMLSVDTV DPID TSLR (Ours) In Figs. 10, 11, and 12, we compare the visual qualities of the denoised images generated by the competing algorithms at noise levels of 30, 50 and 100.…”

Section: Tablementioning

confidence: 99%

“…For complex images, due to the existence of various forms of structures (such as smooth, edge and detailed parts), Elad et al [46] proposed the theory of morphological component analysis (MCA), where the various components of an image are processed separately, and then each component is represented as a sparse combination of atoms of dictionaries, which is applied to image denoising and inpainting. Based on MCA, Shi et al [47] recently proposed a detail preserving image denoising (DPID) method that reconstructs the high-and low-frequency components respectively. In addition, to enhance accuracy, some existing works [48][49][50][51][52] adopted a progressive approach to obtain a rough result and then build a new model.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Two-Stage Image Denoising via an Enhanced Low-Rank Prior

Fan

Shi

et al. 2021

J Sci Comput

Self Cite

View full text Add to dashboard Cite

Restoring the image contaminated with heavy noise remains a challenging task. Since the image prior is essential to restoring a high-quality image, this paper proposes a novel twostage enhanced low-rank prior model (TSLR) for efficient image denoising. Unlike denoising an image as a whole, this algorithm divides the denoising process into two stages: contour restoration and detail restoration. First, we explore the total variation (TV) regularization term to restore the image contour, obtaining the preliminary denoised image. Although TV regularization term can reduce noise, it loses the rich details of the original image. Nevertheless, detail preservation ensures good visual quality of the denoised images. Then, to overcome the above issue, the preliminary denoised image is adopted as a rough evaluation of the original image for the second stage, and the weighted sum of the L 1 -norm and L 2 -norm is utilized as the fidelity term. Furthermore, we introduce a new enhanced low-rank prior, which combines the low-rank prior of similar patches from both gray and gradient domains, to reconstruct the fine details of the image. To further improve the validity of image denoising on the basis of the low-rank prior, the weighted nuclear norm minimization method is adopted in the present study. In addition, this work adaptively selects the search window size for different regions to accurately select similar patches. Through extensive experiments, the results reveal that our scheme can retain more detailed information while eliminating noise and can surpass a variety of advanced non-deep methods regarding both the PSNR and SSIM.

show abstract