Night Image Enhancement
Night images obtained from a surveillance camera have low visibility compared to daytime images. Images captured at night have low brightness, low contrast and high noise. A modified contrast enhancement (CE) algorithm was proposed and developed in luminance-chrominance space. Only the luminance channel obtained by PCA transform is processed as it contains the most valuable information. Daytime images are simulated with various degrees of contrast and Poisson noise using MATLAB. CE algorithm is applied in three scales to obtain good brightness and contrast of the images. Images are denoised by using bilateral filter that smoothes the noise while preserving edges. The brightness and contrast of the night images have been enhanced significantly and the noise is reduced effectively, preserving the details of the images. Finally, the performance of the proposed algorithm is illustrated by processing images under various lighting conditions without introducing halo and ghosting artifacts. Structural similarity and visual contrast measures demonstrate that the proposed method is more effective over other existing methods.
Experiments were performed for different scenes including Indoor and Outdoor environments. Images were captured using DSLR camera and the resolution of the testing images is 1440 X 1080. We compare our results with other methods(contrast pair and denighting). We use SSIM and VCM for quantitative analysis of our results.
Scene 1 - Real night images (Indoor)
Scene 2 - Real night images (Outdoor 1)
Scene 3 - Real night images(Outdoor 2)
1. Steffi Agino Priyanka, Hsiao-Jung Tung, Yuan-Kai Wang, "Contrast Enhancement of Night Images", 16th IEEE conference on Machine learning and Cybernetics, PP 380-385, July 2016.