Research On Nighttime Image Enhancement For Video Surveillance

Nighttime image enhancement for video surveillance is an image processing technology which processes and enhances the images captured by surveillance cameras at the midnight. It tries to figure out the problem in the images(the images captured by surveillance cameras), such as high noise and low-contrast, get a better visual effect and make the human visual system satisfy.This paper summarizes and implements a variety of traditional and recently proposed algorithms for low illumination images. After studying and analyzing those algorithms and theory, the paper propose two algorithms that can effectively increase contrast and weaken the noise by using stochastic resonance theory, affine transform and multiple linear regression. By comparing those algorithms, the proposed ones have a better enhancement effect than the traditional ones.The first algorithm is based on thinking about how to deal with the image’s low illumination and high noise problem. First, we get the low-frequency information of image which reflects the image background and the light intensity by low-pass filter, and then enhance the information by the principle of camera exposure. Second, we weaken the noise and enhance the original image by using multiple iterations of stochastic resonance. Third, we fuse the two part processed information to form a new image, and then we filter the image by morphological filter for a better visual effects.The second algorithm uses an illumination degradation model based on affine transformation to recover the enhanced image from original image. At the same time, we take into account that the illumination of night image is always uneven distribution, that’s because the light source is always uneven distribution and sometimes the light will be blocked, which can’t be solved by affine transformation. Therefore, we apply linear regression to fit the image, figure out a compensated illumination plane and use it to enhance the image whose illumination is uneven distribution.The two algorithms reapply some techniques and principles to process the nighttime image in the surveillance system which are usually not used in this field, and they have a better visual effects than the traditional ones which makes up for the hardware of video surveillance system. Besides, the proposed algorithms in this paper not only can be applied to the nighttime images captured by cell phones or cameras, but also to low-illumination and low-contrast images captured during the day.