Minimize Radar Course 3

Back to Radar Courses

43. Texture and image analysis

The concept of texture is discussed by Laur (1989). Before describing the specificity of radar image texture, it is necessary to define the concept of textural element, i.e. the texture elementary unit, smallest homogeneous element of the same radiometry constituting the texture.

In principle, the texture of a radar image can be described by considering Macro-texture, Meso-texture and Micro-texture. Let us consider the following example to illustrate these three texture categories.

An observer flies with a sensor above a wooded area, high enough not to recognise single leaves but low enough to distinguish clearly tree crowns.
First, he identifies the different stands; these might be oaks, pines, beeches, etc.

The different stands define the macrotexture or structure of the scene. The crowns are the key element to identify the content of a stand and define the mesotexture, a crown being a texture unit in this example.

The grey level of each resolution cell depends on the quantity and orientation of the leaves that scatter the incoming light. Horizontal leaves reflect much more than vertical ones.

This difference between resolution cells of the same class appears in the image as microtexture, called speckle or image noise on a radar image.

The texture of a radar image can be divided into three components:
Micro-texture, i.e. speckle, that appears as grains of the same size as or larger than the resolution cell, and having a random brightness. This texture is inherent to the radar system; it does not correspond to a real variation from one resolution cell to another.

Thus, speckle is essentially an image texture arising from the system, and not from the scene. Speckle degrades image readibility. However, despite this major
disadvantage, speckle may be statistically characterised. This point may be exploited by speckle filtering methods.

Meso-texture or "scene texture" is the natural variation of average radar backscatter on a scale of several resolution cells or more. Taking forests as an example, the high backscatter from the part of the tree facing the radar appears near the shadow of the opposed part of the tree away from the radar.

The result is a grainy texture whose elementary unit covers several resolution cells (depending on the spatial resolution of the system). It is this component of image texture that is most useful in the interpretation of the radar image.

Macro-texture corresponds to variations in radar brightness that extend over many resolution cells. It can be, for example, field boundaries, forest shadows, roads or geologic lineaments.

The structure parameter is extremely important for radar imagery interpretation, especially in geology and oceanography. It is generally assessed using edge or other pattern detection techniques.

Based on this discussion, some observations may be made:
- speckle is superimposed on scene texture and structure, creating problems during textural analysis or edge detection of imagery. In general, speckle is a product function: it is multiplicative, not additive although speckle modelling for scenes of rapidly changing contrast is more complex.
- tone, i.e. average value of the mean cross section σ °, is a local concept. It is the spatial variation of tone that provides "texture".
- texture is related to the spatial distribution of resolution cells and depends on the scene, not on the system. Nevertheless, radiometric quality must be relatively constant in the samples where textural measurements are performed.
- texture and structure are limited by the spatial resolution.
- any post-processing of a SAR image (filtering, texture analysis) should be performed on radiometrically original images (slant range). Ground range images are resampled and thus may be radiometrically distorted.