Image-based fusion for video enhancement of night-time Surveillance
Y.Bao, W.Lin & L.Chen, Optical Engineering letters, dec 2010, vol.49.
This paper deals with the fusion of high quality day and night time images to get enhanced background for night images and to enhance the quality of night time videos. The moving objects are also enhanced by the proposed method called region fusion for improving the sharpness of the moving objects.
First, the image is decoupled into color and intensity layer, where the algorithm is applied in the intensity layer. Then the image is separated into illumination and reflectance component by retinex theory. The illumination component contains the low frequency components and reflectance component contains the high frequency components.
To enhance the night time background, weighted average image fusion algorithm is used in illumination component of the images. Due to low contrast , the moving objects cannot be extracted clearly from the night background. Tone mapping is used to enhance the video frames and separate image into details and large scale features. After enhancing the night time videos , motion detection is done to extract the foreground moving objects, where each pixel is modeled by the mixture of gaussian distributions.
The final video is enhanced by adding the motion detected video frame, night time illumination image and the enhanced background image.
First, the image is decoupled into color and intensity layer, where the algorithm is applied in the intensity layer. Then the image is separated into illumination and reflectance component by retinex theory. The illumination component contains the low frequency components and reflectance component contains the high frequency components.
To enhance the night time background, weighted average image fusion algorithm is used in illumination component of the images. Due to low contrast , the moving objects cannot be extracted clearly from the night background. Tone mapping is used to enhance the video frames and separate image into details and large scale features. After enhancing the night time videos , motion detection is done to extract the foreground moving objects, where each pixel is modeled by the mixture of gaussian distributions.
The final video is enhanced by adding the motion detected video frame, night time illumination image and the enhanced background image.