A LogTVSCAD Nonconvex Regularization Model for Image Deblurring in the Presence of Impulse Noise

Table 1

Performance comparison of different methods for SP and RV.

Image	Level	Method	Motion			Average			Gaussian
			CPU	SNR	SSIM	CPU	SNR	SSIM	CPU	SNR	SSIM
House	0.6 (SP)	TVL1	5.1875	9.3729	0.7272	4.9688	10.8485	0.7484	5.4845	5.3156	0.6282
		TVSCAD	46.8756	15.9418	0.8442	47.1563	16.7165	0.8489	51.8125	16.6146	0.8465
		TVLog	47.0468	15.9401	0.8443	46.7344	16.6897	0.8487	43.4843	16.5666	0.8451
		LogTVSCAD	46.2187	15.9501	0.8443	46.5625	16.7233	0.8490	48.1562	16.6208	0.8466
House	0.9 (SP)	TVL1	5.2344	3.1335	0.6315	5.2031	5.2470	0.6346	5.4845	5.3156	0.6282
		TVSCAD	47.6563	10.8527	0.7644	46.1406	13.2556	0.7943	48.9375	13.360	0.7935
		TVLog	46.9375	9.9173	0.7551	45.2187	11.3900	0.7709	45.9063	10.6551	0.7661
		LogTVSCAD	47.3437	11.7085	0.7655	44.7812	13.4710	0.7972	44.6406	13.5232	0.7958
House	0.7 (RV)	TVL1	5.2969	6.7084	0.6603	5.6406	6.9738	0.6760	5.1093	7.2092	0.6770
		TVSCAD	51.906	11.4798	0.7671	50.3906	13.0353	0.7946	44.0625	13.2004	0.8120
		TVLog	49.75	12.2186	0.7971	50.8593	14.2516	0.8192	44.0469	13.2004	0.8232
		LogTVSCAD	50.2188	13.4190	0.8026	51.0937	14.8019	0.8210	44.3906	15.3713	0.8287
Peppers	0.9 (SP)	TVL1	30.3750	6.9405	0.6093	31.5937	5.5903	0.6282	30.6093	6.6347	0.6255
		TVSCAD	146.6250	12.8617	0.7461	146.1719	11.2206	0.7616	148.3593	12.3083	0.7656
		TVLog	145.5781	13.0023	0.7621	146.8437	12.4857	0.7755	148.0625	12.2156	0.7723
		LogTVSCAD	147.9843	13.2269	0.7674	148.5468	12.9662	0.7756	148.9062	12.7609	0.7746
Peppers	0.7 (RV)	TVL1	29.9843	6.6032	0.6290	30.0312	6.6501	0.6433	30.4218	6.6491	0.6453
		TVSCAD	284.875	13.4355	0.7837	285.7343	13.4512	0.7904	286.5781	13.5209	0.7924
		TVLog	297.7812	13.6295	0.8039	304.6406	13.8962	0.7925	288.9062	13.8775	0.7921
		LogTVSCAD	286.3437	14.3542	0.8043	290.0937	15.1048	0.8130	287.875	15.4038	0.8136

The bold values indicate that the SNR and SSIM results obtained by LogTVSCAD are better than the other three methods.