Understanding shape and tonal characteristics is part of the picture, but there are some other things about exposure that can be garnered from a histogram that are related to these characteristics. Remember, a histogram is merely a guide. The best way to understand an image is to look at the image itself, not just the histogram.
Contrast is the difference in brightness between elements of an image, and can determine how dull or crisp an image appears with respect to intensity values. Note that the contrast described here is luminance or tonal contrast, as opposed to colour contrast. Contrast is represented as a combination of the range of intensity values within an image and the difference between the maximum and minimum pixel values. A well contrasted image typically makes use of the entire gamut of n intensity values from 0..n-1.
Image contrast is often described in terms of low and high contrast. If the difference between the lightest and darkest regions of an image is broad, e.g. if the highlights are bright, and the shadows very dark, then the image is high contrast. If an image’s tonal range is based more on gray tones, then the image is considered to have a low contrast. In between there are infinite combinations, and histograms where there is no distinguishable pattern. Figure 1 shows an example of low and high contrast on a grayscale image.
The histogram of a high contrast image will have bright whites, dark blacks, and a good amount of mid-tones. It can often be identified by edges that appear very distinct. A low-contrast image has little in the way of tonal contrast. It will have a lot of regions that should be white but are off-white, and black regions that are gray. A low contrast image often has a histogram that appears as a compact band of intensities, with other intensity regions completely unoccupied. Low contrast images often exist in the midtones, but can also appear biased to the shadows or highlights. Figure 2 shows images with low and high contrast, and one which sits midway between the two.
Sometimes an image will exhibit a global contrast which is different to the contrast found in different regions within the image. The example in Figure 3 shows the lack of contrast in an aerial photograph. The image histogram shows an image with medium contrast, yet if the image were divided into two sub-images, both would exhibit low-contrast.
A digital sensor is much more limited than the human eye in its ability to gather information from a scene that contains both very bright, and very dark regions, i.e. a broad dynamic range. A camera may try to create an image that is exposed to the widest possible range of lights and darks in a scene. Because of limited dynamic range, a sensor might leave the image with pitch-black shadows, or pure white highlights. This may signify that the image contains clipping.
Clipping represents the loss of data from that region of the image. For example a spike on the very left edge of a histogram may suggest the image contains some shadow clipping. Conversely, a spike on the very right edge suggests highlight clipping. Clipping means that the full extent of tonal data is not present in an image (or in actually was never acquired). Highlight clipping occurs when exposure is pushed a little too far, e.g. outdoor scenes where the sky is overcast – the white clouds can become overexposed. Similarly, shadow clipping means a region in an image is underexposed,
In regions that suffer from clipping, it is very hard to recover information.
Some describe the idea of clipping as “hitting the edge of the histogram, and climbing vertically”. In reality, not all histograms exhibiting this tonal cliff may be bad images. For example images taken against a pure white background are purposely exposed to produce these effects. Examples of images with and without clipping are shown in Figure 5.
Are both forms of clipping equally bad, or is one worse than the other? From experience, highlight clipping is far worse. That is because it is often possible to recover at least some detail from shadow clipping. On the other hand, no amount of post-processing will pull details from regions of highlight-clipping in an image.