High-quality underexposed enhancement performance is highly dependent upon the accurate estimation of illumination. This paper presents an underexposed image illumination map estimation method via a sparse bright channel, which is inspired by the observation that the bright inverted channel of an underexposed image is less sparse. First, we demonstrate this phenomenon mathematically and then validate it experimentally. We also propose an algorithm that enforces the inverted bright channel's sparsity, indirectly estimating a coarse but appropriate initial illumination map. Furthermore, the initial illumination map is refined into a piecewise smoothing and structure-preserving illumination map using an updated weight with joint local exposure and detailed feedback constraints. The enhancement can be obtained accordingly with a well-constructed illumination map while avoiding visual artifacts (\textit{i.e.}, loss of details, color cast, and uneven exposure). Notably, computer simulations show that the proposed model is competitive with or even outperforms several state-of-the-art underexposed image enhancement methods in terms of subjective and objective evaluations.
banzheshitou/sparseBCP
Illumination Map Estimation via A Sparse Bright Channel for Enhancing Underexposed Images