Hierarchical tiling method for identifying surface type in digital images

Context

When working with collections of Synthetic Aperture Radar (SAR) images, statistical modelling-based classification algorithms typically parameterize a distribution function of backscatter values related to surface water in order to assign all of a scene’s pixels to one of the two semantic classes: ‘water’ and ‘no water’. The difficulty in parameterizing such a distribution function originates from the fact that flooded areas often represent only a small fraction of an entire SAR scene. In these circumstances, the distribution of a SAR scene’s backscatter values is often not clearly bimodal and it becomes difficult, if not impossible, to accurately parameterize the distribution function of backscatter values associated with surface water. A sufficiently high percentage of flood pixels is typically required to estimate a reliable and robust distribution function that can be used for accurately delineating all water bodies.

Overview

LIST has developed a scientific algorithm to automatically, efficiently, reliably and accurately delineate water bodies from collections of Synthetic Aperture Radar (SAR) images. The method has three steps. First, the probability density function of the open water backscatter values in the SAR data is estimated. This requires identification of the bimodal aspect with a histogram of backscatter values so that ‘open water’ values can be recognized from other backscatter values. The algorithm uses a hierarchical split based tiling approach (HSBA) that searches for tiles of variable size to make it possible to parameterize the statistical distribution function attributed to surface water-related backscatter values. The tiling and the parameterization processes are fully integrated. The HSBA sequentially splits the SAR image into sub-images of decreasing size in order to identify tiles of variable size for which a surface water-related distribution function can be parameterized.

A theoretical probability density function of water backscatter is then fitted to the histogram of these tiles using nonlinear regression techniques. The backscatter value where this pdf starts to diverge from the histogram is identified. Then isolating those pixels with backscatter values lower than this threshold produces a preliminary flood map that represents the seed region for a subsequent region growing process. This step progressively enlarges the flooded area by including pixels exhibiting a backscatter behaviour that is expected from flooded areas. This method automatically delineates all water bodies in the SAR image(s).

Benefits

• automatic and objective image classification  
• efficient, robust and accurate
• independent of the technical characteristics of the SAR scene
• can be used to estimate the likelihood of flooding for each pixel

Potential Applications

This image processing method can be used to:

• detect different types of changes between two image acquisitions  (e.g. floods)
• detect changed pixels through integration into commercial image processing software
• provide rapid flood extent mapping in case of emergency
• process collections of archived Earth Observation data to generate catalogues of historic flood events and to map flood hazard and risk at large scale