Research On Low-illumination Image Enhancement Algorithm

In the digital image processing, we don't consider the cause of the image degradation when we enhance. According to the image feature and processing intent, we adopt a certain technology to highlight the feature of interest. Meanwhile, it maybe suppress some other information. Finally, it can improve visual effect, or be convenient to analyze and process for human or computer. Its main purpose is to improve the intelligibility threshold.The traditional image enhancement method can be devided into space domain and frequency domain methods. The space domain method is directly to do the calculation for the pixel value. The common space domain methos include liner gray transformation, histogram correction,spatial smoothing and sharpening,pseudo-color processing, etc. Frequency domain method is to enhance the value of the transformation in certain transform domain, and then inverse transform to obtain the enhanced image.As the traditional enhancement methods are simple and obvious effect,they have been widely used in practice.First,this paper for common image enhancement based on histogram equalization algorithm is studied and compared, and then analyzes of its deficiencies that exists in image enhancement.last, It lists several improved methods,combines with experimental results and data,pointing out its strengths and areas for further inprovement.Then,this paper propose a method pixel-based statistics adaptive image enhancement for common fixed surveillance video images. This approach to the relatively simple scenes can get better results. But it can be greatly impacted by a priori image during the day. To this end,the paper finally the basis of previous work proposes an improved denoising filter based on spatial-time accumulation and tone mapping function.It doesn't rely on prior image during day,fully utilizes the information of adjacent frames to enhance the current pixel.On the above two image enhancement methods,this paper provides the process of its implementation algorithm,and a lot of experiments to verify. This improved method for some images can get better results than traditional method.