PEWA: Patch-based Exponentially Weighted Aggregation for image denoising
Authors: Charles Kervrann
NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluated the PEWA algorithm on 25 natural images showing natural, man-made, indoor and outdoor scenes (see Fig. 1). Each original image was corrupted with white Gaussian noise with zero mean and variance σ2. In our experiments, the best results are obtained with n = 7 7 patches and L = 4 images ul denoised with DCT-based transform [26] ; we consider three different DCT shrinkage thresholds: 1.25σ, 1.5σ and 1.75σ to improve the PSNR of 1 to 6 db at most, depending on σ and images (see Figs. 2-3). The fourth image is the noisy input image itself. We evaluated the algorithm with a larger number L of denoised images and the quality drops by 0.1 db to 0.3 db, which is visually imperceptible. Increasing L suggest also to considering more than 1000 samples since the space of candidate patches is larger. The prior neighborhood size corresponds to a disk of radius τ = 7 pixels but it can be smaller. Performances of PEWA and other methods are quantified in terms of PSNR values for several noise levels (see Tables 1-3). |
| Researcher Affiliation | Academia | Charles Kervrann Inria Rennes Bretagne Atlantique Serpico Project-Team Campus Universitaire de Beaulieu, 35 042 Rennes Cedex, France charles.kervrann@inria.fr |
| Pseudocode | No | The paper describes the computational steps but does not include a clearly labeled 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | No | The paper provides links to implementations of third-party methods (BM3D, NL-Bayes, DCT, NL-means) but does not provide a link or explicit statement about the availability of the PEWA algorithm's source code. |
| Open Datasets | Yes | We evaluated the PEWA algorithm on 25 natural images showing natural, man-made, indoor and outdoor scenes (see Fig. 1). Set of 25 tested images. Top left: images from the BM3D website (cs.tut.fi/ foi/GCFBM3D/) ; Bottom left: images from IPOL (ipol.im); Right: images from the Berkeley segmentation database (eecs.berkeley.edu/Research/Projects/CS/ vision/bsds/). |
| Dataset Splits | No | The paper does not provide specific details on dataset splits (e.g., percentages, counts, or predefined partitions) for training, validation, and testing. |
| Hardware Specification | Yes | denoising a 512 512 grayscale image with an unoptimized implementation of our method in C++ take about 2 mins (Intel Core i7 64-bit CPU 2.4 Ghz). |
| Software Dependencies | No | The paper mentions implementation in 'C++' but does not specify any software libraries, dependencies, or their version numbers. |
| Experiment Setup | Yes | In our experiments, the best results are obtained with n = 7 7 patches and L = 4 images ul denoised with DCT-based transform [26] ; we consider three different DCT shrinkage thresholds: 1.25σ, 1.5σ and 1.75σ... The prior neighborhood size corresponds to a disk of radius τ = 7 pixels... In our implementation, we set T 1000 and Tb = 250... β = 4σ2... In practical imaging, we use the method described in [11] to estimate the noise variance σ2 for real-world noisy images. |