Total Variation Regularization Algorithms for Images Corrupted with Different Noise Models: A Review

Figure 5

TV denoising for the mixed Gaussian-Impulse noise model via [118] (a spatially adaptive algorithm).

(a) Grayscale “Peppers” image corrupted with Gaussian plus random impulse noise (𝜎=15/255 and 𝑃=0.3). SNR: 0.98 dB, SSIM: 0.10

(b) Color “Lena” image corrupted with with Gaussian plus Salt & Pepper noise (𝜎=15/255 and 𝑃=0.7)). SNR: −6.0 dB

(c) Color “Lena” image corrupted with with Gaussian plus random impulse noise (𝜎=10/255 and 𝑃=0.2). SNR: 2.72 dB

(d) TV denoising. SNR: 15.09 dB, SSIM: 0.73

(e) TV denoising. SNR: 15.11 dB

(f) TV denoising. SNR: 13.70 dB