ITC analysis of aerial images

• Introduction
• Aerial Sensor Technology
• Image needs for computer image analysis
• Synergy with LiDAR
• BRDF Correction and Normalization of Aerial Data Area
• Tests of feature-based BRDF correction curves on two flight lines
• Conclusion

---

BRDF Correction and Normalization of Aerial Data Area

Relative to satellite images (e.g., IKONOS, QuickBird, etc.), images from aerial sensors bring many additional challenges to an ITC analysis: a) their large view angles (±32°) show increasingly leaning trees as one gets away from the image centre (nadir), making crown delineation difficult and increasing the probability of trees being completely hidden, b) the effects of solar illumination are quite different from one side of the image to the other, as the trees on one side are generally seen as front lit, while on the other side, mostly backlit, c) there is a need to normalize between images and flight lines.

With scanned photos and digital frame cameras, these effects occur in a circular pattern. For “push-broom” sensors, that acquire their images line by line, this occurs in only one direction, across the image. To simplify the discussion, we will address the later case.

Good ITC analyses of aerial data can be obtained by concentrating on the more central regions of images and correcting for (and normalizing) intra- and inter-image variations of vegetation radiances. This can be accomplished via BRDF (Bidirectional Reflectance Distribution Function) correction curves. For multispectral images, different BRDF correction curves are collected and applied to each spectral band.

The first step consists in establishing a curve parameterizing the view angle and sun illumination effects across the image by accumulating a histogram of pixel averages in each column of an image. For this to work properly, one as to assume that the image is very long and/or that the ground features are random enough so that their spatial distribution does not overly affect the curve. This is rarely the case. Alternatively, a BRDF curve can be created from a single feature prevalent throughout the image, preferably the type of feature one is analysing (e.g., forest areas). With high resolution data, experience has shown that it may be better to acquire a separate curve for each specific feature, such as coniferous or deciduous tree crowns, as they respond differently to sun-view angle geometry because of their different shapes (i.e., conical trees vs rounded trees).

The second step consists in normalizing that BRDF curve relative to the grey-level value at nadir. Then, the curve represents typical additive or subtractive values found at off-nadir positions, relative to the average grey-level at nadir. The inversion of that curve becomes the correction curve. It describes the grey-level correction values to apply to each pixel based on its position off-nadir.

BRDF correction curves can also be used to normalize radiances between images and/or flight lines. Because these curves are relative to the average grey-level at nadir, it suffice to decide on a single fixed grey-level value (per channel) to be assigned to the nadir position of all images to normalize the radiances between them.

---