Image Interpretation

Image interpretation is the process used in remote sensing to extract meaningful information from photographic and digital images taken of the Earth's surface and atmosphere. The goal is to identify, count, measure, and classify geographic objects or features in the landscape as they are represented by single images or to infer physical geographic processes from two or more images. Whether these objects are individual trees, clusters of trees, or a whole forest or whether the objects are buildings, neighborhoods, or a complete town depends on the spatial resolution of the remote sensor. Similarly, whether the tree is deciduous or coniferous or whether the building is a house or a store partly depends on the spectral resolution of the remote sensor. Spatial, spectral (and radiometric, referring to gray levels) resolutions refer to the specific designs of the optical-based or pulse-based sensors used in remote sensing, where variations of these resolutions determine the geographic scale and level of detail and the quality of interpretability. For instance, the widely used U.S.-based Landsat Thematic Mapper series has sensors designed with spatial resolutions of 30 m (meters) and spectral resolutions of six or seven bands or channels (referring to energy wavelengths), which makes them ideal for applications where objects are no smaller than 30 m on the ground, while the more recent Ikonos sensors are capable of much more detailed applications where objects of approximately 4 or 1 m should be recognizable.

On close inspection, raw aerial photographs and digital images are found to be composed of intricate patterns of basic tones and textures. Figure 1 illustrates the image interpretation process where these patterns are converted into meaningful geographic information. Most image interpretation procedures require the interaction of human skill, experience, and knowledge, with support from rigorous, systematic computer calculations.

At this point, the interpretation process can follow one of two directions. First is the traditional visual approach used when handling aerial photographs, which involves the systematic search for clues in how target objects on the ground [Page 1535]appear on the photos by understanding the interplay of tone, texture, shadow, pattern, association, shape, size, and site (refer to basic texts such as Campbell, 2006, for detailed discussions of each). The manual approach is heavily dependent on overlapping photos taken along the flight path, which facilitate a three-dimensional (3D) perspective when viewed with stereoscopes. The 3D view aids in object recognition and further assists with the calculation of scale, horizontal distances, and vertical heights of objects.

Figure 1 The image interpretation process

Source: Victor Mesev.

The other approach to interpretation is the computer analysis of digital images, where a number of algorithms are designed to enhance, contrast, and classify target objects based on numeric multispectral similarities. One of the simplest techniques to help identify geographic features is to enhance the visual appearance of the digital image by stretching the brightness levels of the data model used to represent the image. Remembering that raw images replicate the typically homogeneous nature of reality in small areas, pixel values are clumped within narrow data ranges, rendering the identification of geographic features difficult. By stretching this narrow range across the full extent of the data model (i.e., 256 brightness levels for 8-bit words), enhancement methods, such as linear contrast or histogram equalization, are able to “illuminate” and reveal much more detail in the landscape. However, enhancements are for visual purposes only and do not alter pixel values. Alternatively, edge enhancement is a filtering technique that involves a moving window designed to attenuate the values of pixels representing boundaries between geographic objects. This should make the interpretation of objects and features a little easier, although the image remains numeric.

...