Contrast Enhancement

Most aerial photography covered in this book shows landscapes of all sorts, and few of them abound in striking colors with lively contrasts. Even more than high-oblique images, vertical aerial photographs for geoscientific applications tend to be restricted to shades of green, brown, and beige. Yet exactly the subtle differences in color shades may be interesting for distinguishing certain soil properties or vegetation types.

Various ways exist for manipulating the distribution of tones across the brightness range between 0 and 255 (for an 8-bit image), all of which modify the image histogram that shows the relative frequency of occurrence plotted against pixel values. Depending on the software used, the following methods might be available (e.g., Richards and Jia, 2006; Langford and Bilissi, 2007).

• Simple global brightness and contrast controls.

• Advanced controls for specific color and brightness ranges.

• Histogram adjustments using levels.

• Histogram adjustments using curves/graphic functions.

Color enhancements may be done in any of the available color spaces (see Chapter 5.2.5). For example, instead of manipulating the three primary colors red, green, and blue, the saturation component of the IHS color space may be used for slight enhancement of dull colors due to backlit situations. Using the intensity and saturation components of the IHS system would not change the hues in the image, while enhancements in the RGB space are able to produce complex color shifts.

Consider the view of a network of erosion rills on a river terrace (Fig. 11-5). This is what the original image looks

FIGURE 11-5 Vertical kite aerial photograph (with image histograms) of rill and gully erosion on a sedimentary river terrace near Foum el Hassane, South Morocco. Taken with a Canon EOS 350D (Digital Rebel XT) with 28 mm Sigma lens, no image processing applied. Field of view ~70 m across; photo by IM, JBR, and M. Seeger, March 2006.

FIGURE 11-5 Vertical kite aerial photograph (with image histograms) of rill and gully erosion on a sedimentary river terrace near Foum el Hassane, South Morocco. Taken with a Canon EOS 350D (Digital Rebel XT) with 28 mm Sigma lens, no image processing applied. Field of view ~70 m across; photo by IM, JBR, and M. Seeger, March 2006.

FIGURE 11-6 Image in previous figure after histogram contrast enhancement. Image processing by IM.

like, shot as RAW and converted to TIF without any additional processing. The monotonous colors of the loamy sediments differ little from the darker hue of the terrace's rock-strewn surface in the upper and lower right. The indirect light of the overcast day renders the image nearly devoid of shadows. The histograms of image values reflect the homogeneous visual impression by a unimodal distribution in the mid-ranges. The slight broadening and shift toward the higher values from blue to red results in the overall brownish image hues. It is obvious that the scene does not exploit the dynamic range of the sensor (and screen or printer) at all; each band covers only 40-50% of the possible brightness values.

Figure 11-6 shows the same image after a contrast enhancement that expanded the histograms between their extremes, slightly saturating the high ranges. The scene is now a bit unnatural with respect to what we would actually see with our own eyes in the field, but it reveals remarkable variations in the seemingly homogeneous sedimentary layers of the terrace. The smooth crusted top surface now has a distinctive pinkish shade where bare of rocks, while the layer immediately underneath and another one farther below appear bright with a turquoise tinge. Where the two main gullies have cut deeper, the sediment is much darker and reddish-brown again. The color differences result from differences in pedogenetic processes, proportions of organic and inorganic material, and grain sizes in the individual layers, which have been deposited during the Late Glacial and Early Holocene (Andres, 1972).

A similar contrast enhancement was applied to this view of sand-sage prairie vegetation (Fig. 11-7). In the original image, only subtle variations in colors and shadows can be seen. Disregarding the bright gravel road and dark shadows of the trees, the brightness values present for the vegetated areas of the scene only covered 25-30% of the 8-bit dynamic range. By stretching them to the full 256-value range, great variations of color, texture, and shape in the vegetation cover become apparent. Individual sagebrush (Artemisia sp.) bushes become quite distinct as pale-green clumps on subtle sand dunes. Between dunes, circular swales with dark green prairie grass are apparent. Orange spots are patches of prairie wildflowers.

In the examples discussed above, the camera sensor's dynamic range was wider than the scene required. Starting with simple linear contrast stretches (min-max or about ±2.5 std. dev.) often yields good results with such images. A more difficult case is the presence of deep shadow in a scene (see also Chapter 4.7). A contrast stretch as applied in Figure 11-6 would deepen them even further, obscuring the shadowed objects still more and increasing their radio-metric difference to the sunlit parts of the image. This might impede further image processing like classification or stereoscopic feature extraction.

In Figure 11-8, the original image of a heavily shadowed scene is compared with two contrast-enhanced versions. Note the strongly bimodal distribution of the original image values shown in Figure 11-8D (gray histogram) and the generally dark appearance of the desert scene. In real life, the scenery was much brighter with dazzling light, but the automatic, non-corrected exposure settings of the camera have resulted in comparative underexposure (see Chapter 6.4.5). The surfaces and objects obscured by the shadow are of the same type as those illuminated by the sun, so the shadows would decrease or even disappear if the two peaks could be moved or even merged together. The colored

FIGURE 11-7 Sand-sage prairie vegetation in sand hills of western Kansas, United States. Kite flyers next to abandoned farmstead at upper right edge of view. Photo taken by JSA and SWA with a Canon PowerShot S70, May 2007; image processing by IM. (A) Original image without processing. (B) Contrast-enhanced version emphasizing variations in vegetation cover. Small pale-green clumps in lower and left sides are sagebrush on dune crests; dark green prairie grass in swales and around farmstead; orange indicates wildflowers in bloom.

FIGURE 11-7 Sand-sage prairie vegetation in sand hills of western Kansas, United States. Kite flyers next to abandoned farmstead at upper right edge of view. Photo taken by JSA and SWA with a Canon PowerShot S70, May 2007; image processing by IM. (A) Original image without processing. (B) Contrast-enhanced version emphasizing variations in vegetation cover. Small pale-green clumps in lower and left sides are sagebrush on dune crests; dark green prairie grass in swales and around farmstead; orange indicates wildflowers in bloom.

232672' 255

r-

-"'"'

i

J

1 1

1

nil 111 JUUniii

.J

284879 255

0

256

FIGURE 11-8 Vertical aerial photograph of gully erosion near Icht, South Morocco. Field of view ~35 m across. Kite aerial photograph by IM, JBR, and M. Seeger, March 2006; image processing by IM. (A) Original image without processing. (B) Piece-wise linear contrast enhancement using the histogram function modification in (D). (C) Non-linear contrast enhancement using various tonal corrections in photo-editing software.

(D) Histograms of the original image (in gray) and the contrast-enhanced image (B) (in color), with modification graphs (see text for explanation).

(E) Histograms of the contrast-enhanced image (C).

237984 255

to

ID

C 0

256

histograms in Figure 11-8D, belonging to the image in Figure 11-8B, are the result of a piecewise contrast enhancement using a graphic histogram modification (see for example Richards and Jia, 2006). Breakpoint a increases the brightness values of the shadow peak to a level corresponding to the mean values of the sun peak. To counteract the resulting tonal compression of the sunlit regions, breakpoint b raises their highest values to the maximum file value 255. Finally, the curve is flattened a little at the depression between the shadow peak and the sun peak by inserting breakpoint c, compressing the original values where few pixel counts exist.

Even less contrast between shadowed and sunlit regions are left in Figure 11-8B (with histograms in Figure 11-8E): The two histogram peaks have moved closer, in the blue channel even melted together. This is the result of a more sophisticated contrast enhancement realized with professional photo-editing software (Adobe Lightroom). Here specific tonal ranges were modified separately in succession, addressing the lights, darks, and shadows individually.

The contrast enhancements presented above were carried out on 8-bit images that are already compressed versions of the originally higher dynamic range the sensor had recorded. Stretching parts of the histogram to greater contrasts results in missing brightness values or jumps between color shades (banding) as can be seen from the shadow peaks in Figure 11-8D. For strong adjustments, this effect may even become visible to the naked eye in the enhanced image. It can be avoided by processing the image in RAW format at higher radiometric resolution (12 or 14 bit) and converting it to an 8-bit file afterward. As computer screens and printers (let alone our eyes) cannot make use of the larger image depth anyway, the reduction to 8 bit would not degrade the image quality, but the banding caused by the contrast stretch would be smoothed out to invisibility. Working with RAW imagery offers many advantages for post-processing that are beyond the scope of this book; for more details the user is referred to the wide choice of manuals on digital image processing (e.g. Langford and Bilissi, 2008).

The examples discussed so far are meant to improve color contrasts throughout the image. In some cases, however, it might be useful to enhance or even overenhance only parts of the image in order to maximize the discrimination of tonal ranges associated with certain types of features. This is illustrated with an agricultural scene (Fig. 11-9), where only the image values associated with the fallow field were stretched in the histogram, followed by a strong saturation boost in the IHS color space. As a result, the assumedly homogeneous field now shows clear distinctions of different zones, from blackish-blue, to yellow-brown, to pink, to light cream colored. Such techniques are useful mapping aids for soil survey and may help to reduce the number of bore holes in field sampling by up to 50% (Fengler, 2007).

Champion Flash Photography

Champion Flash Photography

Here Is How You Can Use Flash Wisely! A Hands-on Guide On Flash Photography For Camera Friendly People!. Learn Flash Photography Essentials By Following Simple Tips.

Get My Free Ebook


Post a comment