Understanding Dynamic Range in Imaging Systems

Earlier in this series, we covered the basics of image sensor technology—from photons to electrons—which springboarded our discussion on three CCD image sensor architectures: full-frame, interline-transfer, and frame-transfer. We then reviewed the essential structural and operational characteristics of CCD image sensors.

The next step in this discussion is understanding the dynamic range of a CCD image sensor, an exploration that begins with a definition of dynamic range. Fortunately, I’ve already written an entire article devoted to the question, What is dynamic range? In that article, I tried to provide a fairly comprehensive explanation that can be applied to a wide variety of engineered systems.

If we limit ourselves to the context of light-sensitive electronics, we can simplify the definition as follows: dynamic range conveys the maximum variation in light intensity that the sensor can record.

But there’s a lot more to this story, and in this article, I’d like to explore the concept of dynamic range as it applies to digital imaging systems. This way, we’ll begin with a more holistic understanding of this issue, and then in the next article, we can examine dynamic range at the semiconductor- and circuit-design level.

 

Dynamic Range in Digital Imaging Systems

This series focuses on CCDs as components in digital cameras, rather than as isolated light-detecting devices, so let’s take a closer look at the meaning of dynamic range in relation to the overall imaging system.

First, I’ll formally introduce the term luminance. This is almost a synonym of “light intensity,” but luminance specifically refers to light reflected by, emitted by, or originating from a particular object or area. A camera responds to luminance, i.e., to the amount of light that reaches the sensor from a given portion of a visual scene.

We can think about dynamic range as the ratio between the luminance that produces white in the resulting image and the luminance that produces black in the resulting image. In a digital system, white corresponds to the highest possible ADC output value, and black corresponds to the lowest possible ADC output value. If we take the commonly used 8-bit grayscale range as an example, white = 255 and black = 0:

An 8-bit grayscale ranging from black (with the pixel value at 0) to white (with the pixel value at 255). 

 

Black and white in a digital monochrome image are very different from “black” and “white” in a physical scene, for three reasons:

• In visual perception, “black” and “white” are associated with color (or lack thereof), whereas a basic light-sensitive device indicates only tone, i.e., a visual representation of measured luminance.
• In nature, true black would be the complete absence of light: zero luminance, as in not a single photon anywhere to be seen. Things can appear black in images even when they are nowhere near black in physical reality.
• In an image, white indicates maximum luminance. In nature, luminance continues to increase; the intensity of light in the universe is not limited by systems—film camera, CCD camera, human eye, etc.—that measure it.

 

A digital monochrome image in which black and white differ from "black" and "white" in the physical scene. 

 

The bottom line here is that the physical world often presents luminance variations that far exceed a camera’s measurement capabilities. Dynamic range tells us how much variation a particular imaging system can capture, and this in turn helps us to understand how well the system can reproduce scenes with high contrast, that is, with large variations in brightness.

 

The Effect of Dynamic Range

Let’s say you’re photographing a scene that includes a deeply shadowed closet and lacy white window curtains illuminated by direct sun. In the first image, the closet looks black—all the interesting detail has been lost. All right, that’s no problem, we just need a longer exposure time. You increase exposure and take another photo. The closet looks good, but now the curtains are completely white—you’ve lost all the detail in a different portion of the image.

This exercise conveys the fundamental limitation associated with an image sensor’s dynamic range. The scene’s ratio of maximum luminance to minimum luminance exceeds the sensor’s dynamic range, and consequently, you can’t retain detail in both the “highlights” (i.e., the bright portions) and the “shadows” (i.e., the dark portions).

 

High-Dynamic-Range (HDR) Imaging

The example above reminds us that the limited dynamic range of an image sensor doesn’t prevent a camera from adequately capturing both highlights and shadows in separate images. All we need to do is take multiple images with different exposure time.

Thus, we could extend the sensor’s interscene (as opposed to intrascene) dynamic range by combining multiple images, and that’s exactly what people do when they create a high-dynamic-range (HDR) image. The HDR procedure involves precisely aligning multiple images of the same scene and creating a composite that incorporates shadow detail from longer-exposure-time images with highlight detail from shorter-exposure-time images.

I consider intrascene dynamic range to be just dynamic range, and I would prefer to designate interscene dynamic range as perceived dynamic range or something like that. Interscene dynamic range can be extended almost endlessly through the use of optical filters, software tricks, and so forth; it doesn’t tell us much about the capabilities of the most important hardware in the system.

 

Conclusion

Dynamic range is a fundamental means of characterizing engineered systems, and imaging systems are no exception. In the next article, we’ll discuss dynamic range as it relates to the structure of a CCD and to external signal-processing circuitry.