Image texture analysis
Texture is the spatial distribution of tones across the pixels of remotely sensed images, providing a measure of tonal variability. Texture analysis or texture mapping is a common method for delineating surface features that cause localized variations in the brightness and other spectral properties of the satellite image, including shadowing. Texture can be used as an important descriptor of ecological systems by approximating vegetation heterogeneity and other ecological indicators. Local variation in spectral properties is measured within a window of a particular size that is passed across the image to assign each pixel a texture value based on the spectral variability of its neighbors.
The images below show an aerial photo of a forested landscape (left) and a texture analysis of the photo (right). The circles identify an area where texture is low (low complexity and variability in spectral composition of neighboring pixels – left circles) and where texture is high (high complexity and heterogeneity – right circles). In this case, the left part of the figure shows an early-successional even-aged stand, and the right part of the figure shows how canopy complexity and heterogeneity increase with successional stage (image analysis performed by Jason Karl from a 2004 NAIP image).
Chica-Olmo & Abarca-Hernández (2000) identify three major categories of texture processing algorithms: structural, spectral and statistical. Structural methods produce texture values by using repetition of primitive patterns with certain rules of placement, but do not handle irregular patterns well. Spectral methods are based on the Fourier transform, a technique for separating an image into its spatial frequency components. Statistical methods include models of statistical properties such as fractal dimension, autocorrelation, and co-occurrence, and are used most frequently in landscape ecology. Gray level co-occurrence matrix (GLCM) texture analysis is one of the more commonly used statistical methods to describe variation in gray scale values in a local area, although it is extremely computationally intensive to calculate. One of the easiest and simplest methods to assess image texture is simply using the standard deviation of neighboring pixels in a moving window analysis across an image. In the case of object-based image analysis, the standard deviation of reflectance values across all the pixels in the object provides a good measure of texture. Texture analysis can also be performed using digitized historical aerial photographs, and therefore can be useful for assessing changes in vegetation communities from historic states.
Texture analysis produces a map delineating the distribution of vegetation with particular textural properties.
Texture analysis has typically been used to characterize forest ecosystems, but some applications to other ecosystems include:
The usefulness of texture analysis will depend on the appropriateness of a particular technique or algorithm to the study site, the level of differentiation of textural properties between disparate vegetation types, and the scale at which textural measurements are taken. See Ferro (1998) for a discussion of scale in texture analysis, and refer to Strand et al. (2008) for an example of texture analysis overestimating juniper cover due to similar textural properties of juniper compared to other vegetation types.
Data inputs vary based on the method of texture analysis used.
Texture analysis requires image processing and statistical/mathematical modeling software.
Mapping the spatial distribution of three distinct textures based on an aerial photograph (source: Zhou 2006)
Three methods of texture analysis of image objects in the Castle Creek area of southwestern Idaho, including standard deviation and two measures from the gray level co-occurrence matrix (GLCM) method (texture analysis performed by Jason Karl from a 2008 Ikonos image).
Aerial photographs showing change in juniper cover from 1946 to 1998 in southwestern Idaho. Left images are aerial photos from the two years (a and d); middle images show juniper crown diameters in white, estimated by wavelet analysis (b and e); and right images show estimated juniper cover in white based on texture analysis (c and f). In this study, texture analysis overestimated juniper cover due to similar textural properties of juniper and other vegetation types (source: Strand et al. 2008).
You must have an account and be logged in to post or reply to the discussion topics below. Click here to login or register for the site.