Fig. 1. Dual-tree complex wavelet transform

Fig. 2. Flowchart of dual-tree complex wavelet threshold denoising

Fig. 3. Multi-directional gradient operator. (a) Horizontal gradient; (b) vertical gradient; (c) positive diagonal gradient; (d) anti angular gradient

Fig. 4. Average PSNR corresponding to parameter C

Fig. 5. Flowchart of dual-tree complex wavelet threshold denoising with edge enhancement

Fig. 6. Experimental images. (a) Barbara; (b) Hill; (c) Mammogram; (d) Xray

Fig. 7. Denoised image Barbara corrupted with noise of intensity σ=30. (a) Barbara with noise of intensity 30; (b) method in Ref.[2](29.76 dB, 92.67%); (c) method in Ref. [8] (25.69 dB, 84.88%); (d) method in Ref. [13] (24.71 dB, 81.04%); (e) method in Ref. [14] (26.11 dB, 85.48%); (f) proposed method (29.79 dB, 92.71%)

Fig. 8. Denoised image Mammogram corrupted with noise of intensity σ=50. (a) Mammogram with noise of intensity 50; (b) method in Ref.[2] (36.33 dB, 85.55%); (c) method in Ref. [8] (33.59 dB, 78.65%); (d) method in Ref. [13] (33.97 dB, 81.60%); (e) method in Ref.[14] (34.32 dB, 78.44%); (f) proposed method (36.95 dB, 86.92%)

Fig. 9. PSNR and SSIM of denoising image Hill corrupted with various intensity noise. (a) PSNR; (b) SSIM

Fig. 10. PSNR and SSIM of denoising image Xray corrupted with various intensity noise. (a) PSNR; (b) SSIM

Fig. 11. PSNR and SSIM of denoising Set12 dataset corrupted with various intensity noise. (a) PSNR; (b) SSIM