Low-light image enhancement based on variational image decomposition

Due to the significant differences in brightness regions in real-world images, existing low-light image enhancement methods may lead to insufficient enhancement in low-light regions or over-enhancement in normal-light regions, as well as color distortions and artifacts. To overcome this drawback, we propose a real-world low-light image enhancement method based on a variational image decomposition model. In our proposed method, we first grayscale and histogram equalize the low-light image. Then, we use the variational image decomposition model to decompose the histogram-equalized grayscale image into cartoon, texture, and high-frequency detail components. Next, we use a Gaussian low-pass filter (GLPF) to remove the noise in the cartoon component, and use a nonlinear stretch function and a gamma function to enhance and compress the texture component and the high-frequency detail component, respectively. We then merge the processed components to obtain a reconstructed grayscale image. Finally, we convert the low-light image from the RGB color space to the HSV color space and recombine the reconstructed grayscale image with the H and S components to obtain the enhanced image after color space conversion. To validate the effectiveness of our proposed method, we carried out both qualitative and quantitative experiments on 5 datasets, and compared it with 14 other low-light image enhancement methods. The results show that our proposed method outperforms most of the low-light image enhancement methods in both qualitative and quantitative performance.