Image above: False color infrared (NBG) image from an infrablue photo taken from a commercial airliner. Scene is looking south near Delta, Alaska at the dendritic erosional pattern of loess covered hills. Reddish patches are birch and aspen trees on the mostly south-facing slopes. North facing slopes are dominated by permafrost and support stunted muskeg and black spruce forests.
The comparison below demonstrates why NDVI is usually computed using a value for the visible red wavelengths instead of blue for each pixel. Red light is not scattered by air and water vapor as much as blue light and makes a crisper image especially when the air is hazy. Infrablue photos use the red channel for infrared light, so must rely on the blue channel for the values for visible light when computing NDVI. The amount of scattering depends on how much air the light passes through, so for aerial imaging the haze effect increases with camera altitude. For oblique aerial photos, this produces a gradient of increasing bluishness with distance. When the blue channel is used for values of visible light in NDVI calculations for oblique aerials, this produces an artifactual gradient of decreasing values with distance. Two-camera systems which use the red channel from an unmodified camera for the visible light data are less subject to this artifact.
The gigapan embedded below can be seen here: http://www.gigapan.com/gigapans/133294/
Scenes above are looking north at the eastern shoreline of Lake Huron. The infrablue camera was a Canon A810 with Rosco #74 filter (no IR block filter). The RGB camera was a Canon S95. The NIR camera was a Canon A495 with a Wratten 87 filter (no IR block filter). The red channel is primarily near infrared light in the infrablue photo, primarily visible red light in the RGB photo, and all near infrared light in the NIR photo (the Wratten 87 filter passes almost no visible light). The NRG image is made from the infrablue photo and is really NBG. By stretching the histogram of the infrablue photo (Photoshop/levels), the range of NDVI values with distance is reduced (the gradient becomes steeper?).