A wavelet denoising approach based on unsupervised learning model

EURASIP Journal on Advances in Signal Processing

Table 1 The PSNR results of the denoised images by different denoising schemes

Image	Method	σ=10	σ=25	σ=50	σ=70	σ=100
Barbara	Wavelet thresholding	31.24	25.60	22.50	21.73	20.23
	TV	30.56	25.31	22.81	22.31	21.72
	K-SVD	34.11	29.57	25.42	23.30	21.87
	BM3D	34.75	30.36	26.95	25.17	23.24
	Proposed approach	34.34	29.80	27.03	25.85	23.63
House	Wavelet thresholding	32.25	27.60	24.57	23.15	21.58
	TV	34.28	30.43	27.21	25.92	24.41
	K-SVD	35.23	31.20	28.08	25.77	23.82
	BM3D	36.36	32.48	29.32	27.52	25.47
	Proposed approach	35.62	31.60	28.40	27.05	25.84
Flinstones	Wavelet thresholding	30.09	24.66	20.72	19.08	17.70
	TV	30.95	25.86	21.63	20.40	18.80
	K-SVD	31.97	27.87	24.33	22.20	19.64
	BM3D	32.27	28.25	24.98	23.20	21.26
	Proposed approach	32.08	28.36	25.35	24.31	22.50
Bridge	Wavelet thresholding	29.81	24.54	21.74	20.80	19.90
	TV	29.55	25.64	22.85	22.05	21.17
	K-SVD	30.87	26.04	23.10	22.10	21.09
	BM3D	31.03	26.15	23.56	22.56	21.57
	Proposed approach	31.25	26.57	23.98	22.87	21.70
Fingerprint	Wavelet thresholding	30.55	25.11	20.86	19.52	17.72
	TV	29.88	25.40	22.90	21.31	18.03
	K-SVD	32.20	27.28	23.28	20.60	18.40
	BM3D	32.19	27.38	24.32	22.84	21.23
	Proposed approach	32.21	27.80	25.18	23.20	21.83

For each test setting, five results are provided: Wavelet thresholding, TV, K-SVD, BM3D, and our proposed model. The highest PSNR values (best denoising results) are given in bold