Table 2.

Quantitative comparison of reconstructions by approximant, DNN-L-3 and DNN-S (for q = 0.5) based on a test set of 500 images. Each entry takes the form’average ± standard deviation’.

	Average PCC ± std.dev		Average PSNR ± std.dev (dB)		Average SSIM ± std.dev
	p = 1	p = 10	p = 1	p = 10	p = 1	p = 10
Approximant fˆ∗	0.148 ± 0.070	0.182 ± 0.086	8.448 ± 4.182	8.465 ± 4.190	0.231 ± 0.111	0.233 ± 0.112
DNN-L output fˆLF	0.812 ± 0.126	0.878 ± 0.083	16.520 ± 2.693	18.439 ± 2.811	0.878 ± 0.088	0.923 ± 0.063
DNN-L-3 output fˆL-3	0.811 ± 0.154	0.879 ± 0.107	16.529 ± 2.549	18.368 ± 2.322	0.875 ± 0.086	0.926 ± 0.094
DNN-S output fˆ (q = 0.5)	0.869 ± 0.112	0.897 ± 0.073	18.600 ± 2.297	19.072 ± 2.271	0.929 ± 0.081	0.935 ± 0.056