Image Denoising And HDR Image Fusion Based On Low Illumination Solid Imaging Systems

Low illumination video image is obtained with more noise in the night with dark environment or under the condition of long distance. In this video, the useful information and noise are mixed together, so that the video feature is not obvious, the clarity is not high. It is difficult to distinguish the target from the background. So we should use digital image processing technology to reduce the influence of noise.At the same time, in the urban environment of night, due to the impact of urban lighting, the image that collected by the photoelectric imaging system will appear "local saturation" phenomenon. At this time, through the traditional image sensor to shoot high dynamic range of the scene, the bright part of the generated image is over exposure and lost many details. The dark part of the generated image is under exposure which can not be resolved. It is impossible to get high dynamic range image of the real scene by using the traditional image sensor. However we can capture a series of different exposure time images which can reflect the real scene. When the exposure time is longer, the captured image can demonstrate the scene details of the low luminance region. Similarly, when the exposure time is shorter, the details in the high brightness region can be performed well in the obtained image. We can fuse the detail informations from different regions with different exposure time to get the high dynamic range image. The HDR(High Dynamic Range) image which can can display the real information of the scene will be displayed on the common display device.In view of this, this paper studies a variety of video image noise reduction algorithms based on time domain, airspace domain and transform domain. Also we implement these algorithms on Matlab and other software. In time domain, because of the strong correlation between video frame and the frame, a recursive weighting algorithm is adopted. By using the least mean square error criterion, the optimal weighting coefficients are obtained by using the least mean square error criterion. In spatial domain, we studies the median filtering algorithm, the mean filtering algorithm, the wiener filtering algorithm, the wavelet threshold filtering algorithm and the improved bilateral filtering algorithm. And we make the subjective and objective quality evaluation of the noise reduction result images which are obtained by the low illumination CMOS. The research results show that the improved bilateral filtering algorithm can get better visual effect. Above all, it can observably retain the edge, texture and other details of the video image.Based on the "local saturation" background of the image from the low illumination solid imaging device, this paper studies a new type of Laplacian Pyramid HDR imaging algorithm. And on the Matlab and other software, we simulate and implement the HDR imaging with different exposure time images which are captured by the low illumination CMOS. In view of the different fusion results produced by different fusion strategies, the subjective evaluation method and objective evaluation method(information entropy, average gradient, standard deviation) were used to evaluate the HDR imaging quality. The research results show that the fusion image detail preserving is more comprehensive based on the type of Laplacian Pyramid HDR imaging algorithm. And the edge of the fusion image is more clear, and the local features of the image are more obvious. Above all, the subjective visual effect of the fusion image is the best.