|
|
||||||||
|
Noise And How To Avoid It |
||||||||
|
Some of this chapter may seem overly technical. I know how annoying it is trying to wrestle with technical terminology when all you're trying to do is take a good picture of Niagara Falls. But, believe me, the technical concepts are, when all is said and done, very simple. And if you can really, really understand these basic concepts, they can have a huge impact on the quality of your photos.
Noise is all around us. Where there is energy, there is noise. Noise is the static on a distant radio station, snow on a TV screen, fluctuations in the voltage across an electrical line, voices in the background at a bar when you're trying to carry on a conversation, hiss on an audio tape. Noise is spurious information that accompanies a signal, but is not part of it. In fact, noise often distorts the signal by interfering with it or interrupting it. Think of the lost word, here and there, in a private conversation in a crowded room, the "snow" that lowers the quality of a television broadcast, the annoying hiss, clicks and pops when a vinyl recording is played. Noise is the bumps in a road, water spots on a window, a cough in the theater. Noise is all around us and it's in our digital cameras. Digital photographs are made up of millions of individual dots. Each of these dots is a pixel (picture element), and each pixel is represented by information about its brightness, color and saturation. Anything that distorts the accuracy of this information is noise. Noise can show up in a digital photograph as unexpected dots of color. We call this kind of noise chrominance noise (or color noise, or chroma noise). In the example that follows, note the ugly red and green dots where they don't belong:
Figure 17-1 Chrominance Noise (100% Crop) Noise can show up as monochromatic clusters that blot out whole sections of an image, clumps that obscure the subject, background and textures, like the yellow blotches in Fig. 17-2.
Figure 17-2 Heavy Noise At ISO 3200 Or it can show up as luminance noise, errors in brightness and shadow (contrast) that add unexpected and unintended textures or broken patterns where the picture should be smooth:
Figure 17-3 Luminance Noise (100% Crop)
Your car's speedometer is a good example. When the indicator points somewhere between 45 and 50 miles per hour, what is your absolute speed? Who knows? It's "somewhere around 47". It could be 47.5 or 46.9 and still be "47". This sloppiness, this variance, is a basic characteristic of analog devices. Analog is a carrier for noise. The delta between 47.5 and 46.9 (using our speedometer example) could be the difference between a sharp, accurate image and a soft off-color one with blotches, colored spots or dimples. Digital, on the other hand, has no noise. The signal is encoded into 1's and 0's. In a digital device, there no 1 1/2, no .56. Just 1 and 0. Binary, strict, and absolutely accurate.
I like to think of the digital camera as having a "front half" and a "back half".
The front half of your digital camera is an analog capture and measurement device. From the lens to the sensor and even somewhat beyond, your camera is all about light. It "captures" the light (and the information it carries) as an analog measuring instrument, not as a digital device. Why does it read light as an analog measuring instrument? Because photons are too hard to count. Your local pharmacy employs the same technique. When a pharmacist fills a prescription for pills, nothing gets measured. Each pill is counted, one-by-one, by hand or machine. As a result, the count of the pills in your prescription bottle should always be absolute and accurate. But what if the prescription is a liquid? You don't count liquids. You measure them, you weigh them, you use some device to determine how much of the liquid to put in the bottle. Whether that device is a measuring cup with printed markings on the side or a scale to weigh the liquid, these devices introduce the same kind of sloppiness and error that your speedometer does. How thick are the lines on the measuring cup? Do you fill to the top of the line, the center of the line or just touching the bottom? Is the measuring cup absolutely level? How's your eyesight? Any slight deviation produces errors, no matter how small. Your camera's sensor is similar. It can't count photons (the elemental units of light) one-by-one. They're too small, they're constantly in motion, they move too fast (at the speed of light) and there's just too many of them. It's too hard to distinguish one photon from another. So it channels these tiny photons through apertures (openings) into little cups (light wells), one for each pixel, and measures their cumulative impact on the bottom of the cup, which is made of a light-sensitive material that splits off electrons when hit by a photon. In reality, the sensor doesn't even measure the light itself.. It measures the electrons thrown off, the electrical charge that results from the light hitting the sensor's surface. Think of it as measuring light "by proxy". But when I think of all the opportunities for error, of all the possible imperfections in materials, environmental impact (heat and cold), rounding errors and the loss or corruption of data as it moves along the circuits, it boggles my mind. It's a wonder we get any pictures at all. The Back Half Of Your Camera (Digital) The back half of your camera is digital. It is where the data read out from the sensor gets processed. It is a computer.
The job of the processing components and software is to turn data into information. To turn the analog readouts generated by your sensor into human-comprehensible form: an image. It starts out with a critical electronic component, the analog-to-digital converter The A/D converter takes the analog measurements of electrical charge at each pixel location and turns those readouts into digital information - a data stream of 1s and 0s. Note: In Sony cameras, this is called the Clear RAW data. It is not yet a picture. Just a big data stream. In some cameras, this RAW data can be output to a file (unfortunately, not the H-Series) so that the photographer can make his or her own processing decisions before an actual image is produced, rather than relying on the parameters and algorithms programmed into the camera by some engineer when the camera was designed. As the data continues through the processing circuits, the camera applies a number of software algorithms (including the Bayer Algorithm, which produces the actual colors), resulting in a stream of data that represents your image. That's not the image itself, just the cleaned-up, organized information in a usable format. Once this basic data stream is produced, the back half of the camera really goes into high gear, applying all the preferences you set and its own pre-designed rules to the data stream. The pixel-by-pixel information is actually altered to produce an internal digital image, a bitmap.. Then, though the data is still digital, the camera goes back into a quasi-analog mode to apply corrections, such as white balance, sharpening and noise reduction. Noise reduction is a qualitative process. The camera uses its built-in software to decide what to do about what appears to be incorrect information. This is a very subjective decision as to how much detail to sacrifice toward cleanliness and accuracy, and how to accomplish the goal of noise-reduced images once the parameters are defined. You have little or no control over this process (the H-Series cameras, to date, don't offer any user-defined noise-reduction preference settings), and each camera and each processing engine applies different rules and approaches. The bitmap is converted into a JFIF format which is, in turn, compressed and saved as a JPEG file. JPEG compression is a lossy process. Information is discarded to make the final file smaller and easier to store and manage. High-frequency information is discarded. Detail may once again be lost, artifacts may be created. And if they are, it's just another type of noise. For the H7 and H9, Sony used its branded Bionz processor to handle all this back end work. And though the process is digital, it is not absolute and any number of errors can be introduced by the processing engine. These errors may not be what we normally think of as noise, but can affect the quality of your images, nonetheless. On the other hand, errors produced by increased ISO or contrast, incorrect white balance and over-sharpening can indeed produce classical noise as described earlier in this chapter.
There are many opportunities for noise to be generated throughout the relatively complex chain of components and processes our cameras use to capture a digital image and generate an output file.
During this long, slow march, the data stream is exposed to all kinds of electromagnetic radiation, which causes all sorts of problems. There are motors in your camera (focus, zoom) and magnetic and piezo-electrical devices (image stabilization) that are constantly turning on and off. There are processors. There are all kinds of electrical and electronic components that generate all kinds of fields, creating a miasma of potential interference and distortion that the data stream has to pass through, as intact as possible. Note: Sony has recently pioneered an ingenious solution to this problem. In its newest generation of CMOS processors (Sony a700, Nikon D300), the A/D converter is built into the sensor chip. Not just one A/D converter, but one converter for each column. In a camera with a pixel-density of the H7 and H9, that would mean 3264 A/D converters. Since there are so many, and they're each reading only one column of 2448 pixels (instead of 8.1 million) the data stream is thousands of times smaller and faster than previous chips. The length the data has to travel is much shorter. Because of this, there is much less opportunity for corruption, pollution and data loss. So far, the a700 and the D300 seem to prove that Sony's new approach works, as they have much lower noise levels than the previous generation of CCD sensors despite having more pixels. I'd be willing to bet that the next-generation H-Series cameras have this on-chip A/D converter architecture supporting at least 10 mp of resolution - and with image quality as good as, or better than, current generations. It just proves that packing more pixels into a sensor does not necessarily result in a loss of image quality.
Of course, there may be errors. And, if there are, either you lose some detail or you gain noise, depending on the nature of the error.
Note: This ISO 400 shot also exhibits serious chroma and luminance noise, which may have resulted from the flawed A/D converters. It goes to demonstrate how important the quality of the A/D converters is to producing low-noise images.
It isn't noise itself that causes image-quality issues with digital cameras. It's the visibility of noise. Let's face it: if you can't see the noise in an image, it doesn't matter. All shots in all digital cameras have noise. It is the static that's carried along with the image. But when does that static start to interfere with the image? When does it start to be destructive? This is determined by the signal-to-noise ratio (S/N Ratio). Let's go back to our radio example. Assume you have a tiny bit of static on the station you're listening to. And let's assume you're listening to a symphony. When the orchestra is playing full-out, when the percussion is pounding and the brass is blaring, the static may be inaudible. That's because the signal (in this case, the music) overwhelms the static.
Fig. 17-6 High Signal, Fixed Noise (High S/N Ratio) But in the very quiet passages, when the signal is very low, the static suddenly becomes an annoying counterpoint to the lovely music. That's because the signal-to-noise ratio is low and the noise takes on a much greater importance in relation to the music.
Fig. 17-7 Low Signal, Fixed Noise (Low S/N Ratio) It's exactly the same with digital cameras. The brighter the light, the more light collected by the sensor, the less perceptible the noise is. But when the light is low, when there's plenty of shadow, when a few photons makes a difference between black and gray, the noise can kill detail and smudge up an otherwise lovely picture. There are all kinds of parameters that determine the visibility of noise in digital photography. For instance, if you're printing a high-resolution image at a small size, the dots that comprise the noise may be too small to be visible to the average viewer. If the shadows in your scene are very dark, and there's little detail available in those shadows, the shadow noise may not be noticeable at any size. But if you brighten up those shadows, watch out! The noise may jump right out at the viewer and spoil your shot.
It is generally accepted that smaller sensors have smaller photo sites and produce more noise. They capture less signal, yet the base noise level is approximately the same regardless of the size of the sensor. That's not quite accurate. In fact, it's not the actual sensor size, but the size of the aperture and light well that determine how much signal the sensor can capture. Sony used a breakthrough design on the DSC-H5. By condensing and moving some of the electronics stuffed between each light well, Sony managed to maintain the same aperture and light-well size as the H1 despite having two million more pixels on the identical real estate. It has two million more of the same-size pixels, just packed closer together. And there are differences in the amount of noise generated from camera model to camera model. There are some cameras that are not as well shielded as others, whose A/D converters are of a lesser quality and whose components generate more powerful electro-magnetic fields. But, given the same baseline noise and no dramatic improvements in sensor design, it is generally true that a camera with a smaller pixel will gather less light and have a lower signal-to-noise ratio, resulting in more noise. But more visible noise? That depends on the scene, the light, the size of the print and how the camera processes noise internally.
That's true of any way you increase the signal after-the-fact. Capturing more light, in camera, does not cause more noise, because it increases the signal-to-noise ratio, which is a good thing. But increasing the signal after the light is gathered has a bad side-effect: it increases the noise as well. The signal-to-noise ratio does not improve, but the noise may move over the threshold of visibility. You may end up with too much noise to print at reasonable size, or the noise becomes so obvious that it overwhelms the detail in the image.
The more you amplify the signal, the worse it gets. Try playing that soft section of the radio program with the slight static at a high volume. The noise gets just as loud and much more annoying than it was when the station was played at a low volume. Ever play cassette tapes? Ever turn up the volume? The tape hiss, which was barely noticeable at low volume, can overwhelm the music at high volume. That's exactly what you're doing when you turn up the ISO in a digital camera. Only the best sensors with the highest signal-to-noise ratios produce images that survive radical amplification relatively unscathed, regardless of the size of the sensor. There are many expensive DSLRs with relatively large APS-C size sensors (4-times the size of the Sony H-Series sensors) that still show lots of destructive noise at ISO 800 and above. Yet others that manage the noise better, often through some kind of effective on-chip or software noise reduction that hides the noise more effectively.
Nothing diminishes the risk of noise more than getting the right exposure when the picture is shot. Expose "to the right". Get the brightest picture you can without blowing out your highlights (clipping). That will ensure the highest signal-to-noise ratio and the cleanest possible image for both the light and the camera settings. I can't stress this enough. Proper exposure minimizes noise. Even at higher ISOs, a well-exposed image may be cleaner than a poorly-exposed image at a lower ISO. Raising the ISO can increase the noise in your image, but brightening an underexposed image after it's been shot and saved will produce much more noise than setting the camera's ISO - only because it's so late in the process.
Fig. 17-8 Correct Exposure and Post-Processing (100% Crop) If you look carefully at the bottom half of Fig. 17-8, you'll see a proliferation of red and green color blotches where they shouldn't be. It's winter, the trees aren't green and they're definitely not red. In fact, the entire color caste has been altered (to magenta) by the presence of noise. This noise surfaced when a section of the underexposed image in Fig. 17-12 was brightened up in Photoshop CS3.
How do you know you've got the best S/N ratio possible? By using the H-Series cameras' live histogram to determine exposure. The histogram is a graph that displays information about the luminance of your scene (or your image, after it's shot). The spikes represent the number of pixels at different brightness levels, with the brightest at the right edge, the darkest at the left border. In order to minimize noise, you'll want to look at the right side and make sure that your histogram comes close to or "just touches" the right-hand edge without going over it.
Figure 17-9 Exposures And Histograms Any spikes that go past the right side will be clipped - they won't even show up in the image. Clipping causes highlights to "blow out", resulting in white skies and lost detail on reflective surfaces. The following three pictures were shot in the middle of a snowstorm, just before dusk. These are the same pictures that generated the histograms in the preceding illustration.
Fig 17-10 Overexposed Image - From Top Histogram
Fig 17-11 Correct Exposure - From Middle Histogram
Fig 17-12 Underexposed Image - From Bottom Histogram Keep ISO In Low-Noise Range If you can't get an appropriate exposure by adjusting aperture (f/stop) and shutter speed, then raise the ISO. This is not as good as using just aperture and shutter speed, as ISO is an amplification that can introduce additional noise. All of the H-Series cameras take relatively noise-free images (there may be noise in the darker shadows) from ISO 64 to ISO 200. It's only when you get over that ISO level that you have to start making decisions about whether the shot can be taken with sufficient image quality. ISO 400 is very noisy in the H1, but not bad in the H2-H9 cameras. If the images are properly exposed, these cameras will produce a clean image. If they're underexposed, all bets are off. Any levels above ISO 400 are high-risk. I've shot some concert photos successfully with ISO 800 on the H9 and the H5, but they definitely required post-processing noise-reduction before I could print, even at 5X7. So, if you have to, go ahead and use ISO 800 on these cameras, just be prepared to work on them before you post or print.
The ranges in Table 17-8 represent the low-noise ISO settings available in each camera. You can exceed them, on occasion, and get reasonably good results, but only if that ISO produces an excellent exposure.
If you end up having to correct white balance in post-processing, you're likely to increase noise when the red or blue channel is amplified to re-balance the color. Green is the baseline for color in digital cameras. It carries most of the luminance data - the brightness and contrast that define edges. Therefore, only the brightness of the red and blue channels are altered when you change white balance. If one or both are amplified (brightened), you'll get the same kind of noise you get when you brighten the whole image. So, like exposure, it pays to get white balance right - in the camera.
Noise doesn't matter. Only visible noise matters. If the viewer of your photos can't see the noise, your image is "clean". Which means that you have another option: ignore the noise. If you're sure you're only going to use your images at small size on the Web, or print them at 4X6, you can be more adventurous with ISO and exposure.
One of the reasons I'm not crazy about the H-Series' Vivid Color Mode is that it's applied not just to the information in the image, but also to the noise. When you're using higher ISOs or working with risky exposures, make sure your Color Mode is set to "Normal" or "Natural" otherwise you might just end up with vivid noise instead of high-saturation images.
I post-process all of my photos. Some just get slight changes in levels, some get cropped or rotated, others get minor color adjustments. But they all get sharpened. When you post-process, the order in which you execute operations matters. That order is called your workflow. If you sharpen before you do noise reduction, you'll sharpen the noise. If you increase brightness before noise-reduction, you'll brighten up the noise, making it even more visible. If you increase saturation before noise-reduction, your noise will be more vivid than ever. So here's a general rule of thumb when post-processing your H-Series pictures: reduce noise first, sharpen last. This has always worked well for me.
Noise is an unavoidable part of Digital Photography. You can minimize it through careful settings, correct exposure and judicious use of higher ISOs.
|
||||||||
|
|
