Haze Reduction - Haze Optimized Transform Method

A previous article discussed atmospheric physics and the Dark Area method for haze reduction. This article discusses an alternative method called the Haze Optimization Transform (HOT). The HOT method was developed by Zhang and others from the Canadian Center for Remote Sensing. It is based on the premise that in clear images, the digital numbers (DNs) of the pixels in all the band files are different but highly correlated. As explained in the previous article, the blue band generally exhibits more scattering due to atmospheric haze than the other bands. The presence of haze decreases the correlation between the blue band and the red band, which is much less affected by haze.

Zhang proposed the following: First, prepare a scatter plot of the DNs of the blue band and the DNs of the red band. Determine the best-fit line through the scatter plot using linear regression. Highly correlated pixels will tend to fall on the line. Those altered by light scattering will tend to be less correlated and will fall of the line. Zhang proposed determining the HOT number for each pixel as the perpendicular (or orthogonal) distance from the HOT line. The HOT number can then be used to remove the haze component of each band.

HOT= DN_blue sin(θ) - DN_red cos(θ)

The concept is depicted in the diagram below.

The PANCROMA HOT utility is a three-file method like the Dark Area method. To use it, open your three band files in the order blue, green and red by selecting 'File' | 'Open'. After your files have loaded, select 'Preprocess' | 'Haze Reduction' | 'Haze Optimized Transform'. After a few moments, the haze reduced RGB image will be displayed. Un processed and processed images are shown below

The first image shows considerable cloud and haze. The second (haze optimized transform) image shows significantly improved contrast and detail compared to the unprocessed image. The overall image brightness is better, and the color tones of the unprocessed image have been retained better than in the Dark Areas method shown previously. The images are available at full scale to the right.