Abstract

An approach to remove light cloud cover and atmospheric haze from Landsat multispectral images is described. The algorithm consists of three steps: isolation of regions of degradation; statistical extrapolation; and adaptive filtering. Statistical pattern recognition and region growing techniques are applied to locate regions of the image corrupted by haze and light cloud cover. First- and second-order statistics in neighboring 4 × 4 pixel regions are used as measures of similarity in determining the satisfaction of conditions for region growing. Pixel values within the degraded regions are modified to match the statistic of the surrounding regions using extrapolation techniques. An adaptive filtering algorithm is then applied to the entire image for contrast enhancement.

© 1985 Optical Society of America