Image Denoising Based on Wavelet for IR Images Corrupted by Gaussian, Poisson & Impulse Noises

Summery Image denoising has remained a fundamental problem in various applications of image processing. This paper proposes a new denoising algorithm on Cohen-Daubechies-Feauveau wavelets (CDF 9/7) wavelet transform. We first applied the lifting structure to improve the drawbacks of the wavelet transform where conventional wavelet transforms and other classical decompositions seem to be restricted or limited to handle. Our proposed algorithm in this paper is very efficient in estimating and reducing noises for the contaminated images by the most popular noises such as Gaussian noise, Poisson noise and impulse (salt& pepper) noise. In this algorithm, the noisy image is first decomposed into many levels obtained from different frequency bands and then to be found the best decomposition level for the noise removal. Experimental results on several conditions are investigated for infrared images as study cases under our proposed algorithm. They are very impressive, for example under the noise with σ = 0.2 and density = 20%, for mean square error (MSE) our method decreasing 83%; peak signal to noise ratio (PSNR) increasing 98% and mean of structural similarity (MSSIM) increasing 95%, multi-scale structural similarity (MSSSIM) enhancing 93%, Feature similarity (FSIM) index growing 98.8%, Riesz-transform based Feature Similarity index (RFSIM) increasing 83.4% with the same conditions in other methods. Obviously, the experimental results shown for our proposed algorithm are significantly superior to other related methods.