Noise Estimation and Type Identification in Natural Scene and Medical Images using Deep Learning Approaches

Table 8

Performance of proposed methodology with other authors and algorithms.

Reference no.	Methodology	Noises considered	Dataset used	Accuracy
[29]	ANN	Impulse and electronic	Natural images	93.75%
[28]	CNN with PCA	Gaussian, salt-and-pepper, speckle, and Poisson	Natural images from SIPI image database–misc	99.3%
[14]	Moment-based classification using kurtosis	Impulse noise	Natural images from SIPI image database–misc	Not given
[32]	NN with moments	Gaussian, speckle, salt-and-pepper	Natural images	ANN 87% and NN 90%
[31]	PNN with kurtosis and skewness	Gaussian, speckle, salt-and-pepper and non-Gaussian	Natural images	84%
AdaBoost	AdaBoost	Gaussian, motion, Poisson, salt-and-pepper, and speckle	Medical images	86.93%
Gaussian Naïve Bayes	Naïve Bayes	Gaussian, motion, Poisson, salt-and-pepper, and speckle	Medical images	86.93%
Bernoulli Naïve Bayes	Naïve Bayes	Gaussian, motion, Poisson, salt-and-pepper, and speckle	Medical images	76.83%
Proposed methodology	CNN with HH sub-band	Gaussian, motion artifacts, Poisson, salt-and-pepper, and speckle.	Natural images from SIPI image database–misc	100%
Proposed methodology	CNN with HH sub-band	Gaussian, motion artifacts, Poisson, salt-and-pepper, and speckle.	Medical images of MRI, CT, and ultrasound	96.3%