Denoising of Surveillance Video Using Adaptive Gaussian Mixture Model Based Segmentation Towards Effective Video Parameters Measurement

In recent times, capturization of video became more feasible with the advanced technologies in camera. Those videos get easily contaminated by noise due to the characteristics of image sensors. Surveillance sequences not only have static scenes but also dynamic scenes. Many efforts have been taken to reduce video noise. Averaging the frame as an image had limited denoising effect and resulted in blur. Such result will be avoided if we separate foreground and background, and make background to be averaged only. Recently, a number of video object segmentation algorithms have been discussed and unfortunately most existing segmentation algorithms are not adequate and robust enough to process noisy video sequences. Since the target video contains noise, a large area of background is incorrectly classified as moving objects and obvious segmentation error will appears. Therefore for robust separation, a segmentation algorithm based on Gaussian Mixture Models adaptive to light illuminations, shadow and white balance is proposed here. This segmentation algorithm processes the video with or without noise and sets up adaptive background models based on the characteristics of surveillance video to accomplish segmentation, reducing background noise by averaging and foreground noise by ML3D filter. The proposed method increased PSNR about 4.5 db compared to existing method and is capable of preserving video content. It is performed for two different video sequences.