|
| References | Methodology | Dataset | Performance measure and accuracy | Gaps identified |
|
| [17] | Noise identification and denoising from the grayscale image; Noise: impulse noise | Dataset: Baboon, Cameraman, Lena, peppers, and Pemaquid images (512 × 512 image resolution) | Accuracy: PSNR = 47.278, SSIM = 0.978, SDME = 61.637 | The image size is only five; Identify only noisy pixels |
| [29] | Noises: impulse and electronic; Statistical features: kurtosis and skewness fed to the ANN | Dataset: Kaggle dataset of natural images; Two hundred images of both noises. Training 60%, testing 40% | Accuracy 94.37% | Lesser number of images are considered for training and testing |
| [13] | Noises: Gaussian; Estimate noise level using DWT coefficients | Dataset: MRI images of T1 and T2 weighted | Estimate noise | No classification |
| [30] | Noise: Gaussian, speckle, line pattern stripes, and circle pattern ring; Two-cascaded CNN model is designed | BSD dataset and 1,000 CT dataset from NBIA, 300 SEM dataset from Dartmouth | Performance measure: PSNR and SSIM; PSNR-37.46, SSIM-0.9001 | They are estimating the noise and denoising using CNN |
| [28] | Noises: impulse, Gaussian, speckle, and Poisson; To reduce the computation time, the PCA filters are used | Dataset: natural images from SIPI dataset– misc | We have carried out three experiments. Four types of noise combinations of noises; Overall accuracy 86.3% | They are considered four types of noise impulse, Gaussian, speckle, and Poisson, with eight classifiers |
| [14] | Noise: impulse; DCT to obtain the kurtosis in terms of a sum of absolute deviation to identify impulse noise | Dataset: natural images USC-SIPI Image database; Size 170 images of different noise levels | Accuracy: 97% | The data set is small and natural images. Considers only impulse noise |
| [15] | Noise: Gaussian, Speckle, and salt and pepper; Statistical features such as kurtosis and skewness | Dataset: few images from natural and medical | Features: kurtosis, skewness; Accuracy: not given | Few images |
| [16] | Noises: Gaussian, speckle, salt-and-pepper; Features: kurtosis and skewness. Method: PNN | Dataset: natural images; Size: not mentioned | Accuracy: 82% | NA |
| [31] | Gaussian, speckle, salt-and-pepper; Features: kurtosis and skewness. ANN for classification | Dataset: natural images; Size: 180 images | Accuracy: average 84% | NA |
| [32] | Noises: Gaussian, speckle, salt-and-pepper; Features: kurtosis and skewness. Method: KNN, NN | Dataset: natural images; Size: 70 | Accuracy: ANN: 87%, NN: 90% | Seventy natural images such as cameraman are used |
|
