Segmentation and statistical analysis of ground-based infrared cloudy sky images

Abstract

The statistical behavior of ground-based IR cloudy sky images are analyzed to acquire a priori knowledge of the background necessary for the development of sophisticated infrared target detection and recognition systems. Infrared cloudy sky images can be automatically segmented into their related groups of interest, according to the radiance statistical distribution, by implementing specialized image processing techniques. Once the images have been segmented, the cloud cover, statistical distribution, and other parameters of interest are readily obtained. It was found that, in the 8- to 1 2-μm spectral window, cloudy sky images must be divided into at least five regions of interest, and in the 3- to 5-μm spectral window, a distinction must be made between shaded cloud images and sunlit cloud images. Shaded cloud images have only three regions of interest, whereas sunlit cloud images are more complex and have at least five regions of interest. If each region is approximated by a Gaussian distribution, then the normalized cross-correlation function ofthe measured data with the multinormal function gives a value in excess of 0.9.