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

Figure 5

(a–c) Three underexposed photos shot in the same scene with different exposure levels. (a) Slightly underexposed. (b, c) Relatively severely underexposed. (d–f) Restoration results of a state-of-the-art method DeepUPE [3], which is a learning-based approach. (g–i) Results of LiAR, the approach proposed in this paper. It is not difficult to find that DeepUPE [3] can restore slightly underexposed images very well while performing quite poor for the severely underexposed ones. By contrast, LiAR, as an image-specific approach, can deal with images with a wide range of exposure levels quite well.

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)