Revisit Retinex Theory: Towards a Lightness-Aware Restorer for Underexposed Images

Table 1

Objective evaluation results of the competing methods.

Methods	Pretrained	Group A		Group B		Group C		All Groups
		CDIQA	LOE	CDIQA	LOE	CDIQA	LOE	CDIQA	LOE
HE [12]	NO	2.7240	0.0540	2.9697	0.0478	3.0058	0.0537	2.8998	0.0518
CLAHE [14]	NO	3.2602	0.1352	2.8724	0.0988	2.1903	0.0632	2.7743	0.0991
Retinex [6]	NO	3.1847	0.1566	3.1693	0.1625	3.0282	0.1943	3.1274	0.1711
Yuan and Sun [19]	NO	2.8227	0.3041	3.1804	0.1105	2.3889	0.0354	2.7973	0.1500
LIME [8]	NO	3.2417	0.1955	3.3036	0.1490	2.9276	0.1352	3.1576	0.1599
Exposure [28]	YES	3.2930	0.0295	2.7563	0.0298	2.3231	0.0606	2.7908	0.0400
DeepUPE [3]	YES	3.3312	0.0680	2.7525	0.0294	2.1580	0.0254	2.7472	0.0409
ExCNet [21]	NO	3.0677	0.2893	3.2757	0.1239	2.6697	0.0566	3.0044	0.1566
LiAR	NO	3.2798	0.0324	3.2750	0.0576	2.7801	0.0569	3.1116	0.0490