Adaptive intensity transformation for preserving and recovering details in low light images

Abstract

It is extremely difficult for intelligent systems to work effectively in dark environments due to high noise level and low contrast. To overcome this difficulty many contrast enhancement techniques have been developed to raise intensities of low light images to allow autonomous systems operating under low light conditions to work effectively. However, in today's world of night time autonomous navigation and surveillance, most of images captured by visual sensors are not always completely dark but in many cases contain a mix of dark and bright regions. Especially, in low light urban scenes where street lights and illuminated sign posts form part of the scene. This, however, can be challenging for many image enhancement and night vision algorithms, as it is difficult to find an optimal balance in improving contrast in the dark regions without compromising on detail and contrast in the bright regions of low light images with mixed dark and bright regions. Therefore, we propose an improved intensity transformation method which uses a combination of the log and identity transformations to improve the quality of images captured during low light conditions. The proposed method is adaptive and can adapt to low light images with mostly dark regions, to improve details and contrast, just like traditional methods. It can also adapt to the images with mixed of dark and bright regions by analysing the intensity distribution of the input image to estimate suitable parameters, in order to robustly improve contrast in the dark regions and preserve details in the bright regions of such images. We demonstrate the effectiveness of our proposed method with relevant experiments as well as making comparisons to conventional methods.