Photoshop.CS.Bible [Electronic resources] نسخه متنی

Noise Factors

Photoshop offers four loosely associated filters in its Filter Noise submenu. One filter adds random pixels — known as noise — to an image. The other three, Despeckle, Dust and Scratches, and Median, average the colors of neighboring pixels in ways that theoretically remove noise from poorly scanned images. But in fact, they function nearly as well at removing essential detail as they do at removing extraneous noise. In the following sections, I show you how the Noise filters work, demonstrate a few of my favorite applications, and leave you to draw your own conclusions.

Adding noise

Noise adds grit and texture to an image. Noise makes an image look like you shot it in New York on the Lower East Side and were lucky to get the photo at all because someone was throwing sand in your face as you sped away in your chauffeur-driven, jet-black Maserati Bora, hammering away at the shutter release. In reality, of course, a guy over at Sears shot the photo for you, after which you toodled around in your minivan trying to find a store that sold day-old bread. But that’s the beauty of Noise. It makes you look cool, even when you aren’t.

You add noise by choosing Filter Noise Add Noise. Shown in Figure 10-41, the Add Noise dialog box features the following options:

Figure 10-41: The Add Noise dialog box asks you to specify the amount and variety of noise you want to add to the selection.

Amount: This value determines how far pixels in the image can stray from their current colors. The value represents a color range rather than a brightness range and is expressed as a percentage. You can enter a value as high as 400 percent. The percentage is based on 256 brightness values per channel if you’re working with a 24-bit image and 32,768 brightness values for 16-bit images. So with a 24-bit image (8-bit channels), the default value of 12.5 percent is equivalent to 32 brightness levels, which is 12.5 percent of 256.

For example, if you enter a value of 12.5 percent for a 24-bit image, Photoshop can apply any color that is 32 shades more or less red, more or less green, and more or less blue than the current color. If you enter 400 percent, Photoshop theoretically can go 1,024 brightness values lighter or darker. But that results in colors that are out of range; therefore, they get clipped to black or white. The result is higher contrast inside the noise pixels.

Uniform: Select this option to apply colors absolutely randomly within the specified range. Photoshop is no more likely to apply one color within the range than another, thus resulting in an even color distribution.

Gaussian: When you select this option, you instruct Photoshop to prioritize colors along the Gaussian distribution curve. The effect is that most colors added by the filter either closely resemble the original colors or push the boundaries of the specified range. In other words, this option results in more light and dark pixels, thus producing a more pronounced effect.

Monochromatic: When working on a full-color image, the Add Noise filter distributes pixels randomly throughout the different color channels. However, when you select the Monochrome check box, Photoshop distributes the noise in the same manner in all channels. The result is grayscale noise. (This option does not affect grayscale images; the noise can’t get any more grayscale than it already is.)

Figure 10-42 compares three applications of Gaussian noise to identical amounts of Uniform noise.

Figure 10-42: The Gaussian option produces more pronounced effects than the Uniform option at identical Amount values.

Noise variations

Normally, the Add Noise filter adds both lighter and darker pixels to an image. If you prefer, however, you can limit the effect of the filter to strictly lighter or darker pixels. To do so, apply the Add Noise filter, and then apply the Fade command (Ctrl+Shift+F on the PC or z -Shift-F on the Mac) and select the Lighten or Darken blend mode. Or you can copy the image to a new layer, apply the filter, and merge the filtered image with the underlying original.

Figure 10-43 shows sample applications of lighter and darker noise. After copying the image to a separate layer, I applied the Add Noise filter with an Amount value of 40 percent and selected Gaussian. To create the upper-left example in the figure, I selected Lighten from the blend mode pop-up menu. To create the upper-right example, I selected the Darken mode. In each case, I added a layer of strictly lighter or darker noise while at the same time retaining the clarity of the original image.

Figure 10-43: You can limit the Add Noise filter to strictly lighter (left) or darker (right) noise by applying the filter to a layered clone. To create the rainy and scraped effects (bottom examples), I applied Motion Blur and Unsharp Mask to the noise layers.

To achieve the streaked noise effects in the bottom examples of Figure 10-43, I applied Motion Blur and Unsharp Mask to the layered images. In the Motion Blur dialog box, I set the Angle value to –30 degrees and the Distance to 30 pixels. Then I applied Unsharp Mask with an Amount value of 200 percent and a Radius of 1. Naturally, the Threshold value was 0.

Chunky noise

My biggest frustration with the Add Noise filter is that you can’t specify the size of individual specks of noise. No matter how you cut it, noise only comes in 1-pixel squares. It may occur to you that you can enlarge the noise dots in a layer by applying the Maximum or Minimum filter. But in practice, doing so simply fills in the image, because there isn’t sufficient space between the noise pixels to accommodate the larger dot sizes.

Luckily, Photoshop provides several alternatives. One is the Pointillize filter, which adds variable-sized dots and then colors those dots in keeping with the original colors in the image. Though Pointillize lacks the random quality of the Add Noise filter, you can use it to add texture to an image.

To create the top-left image in Chapter 13.)

Figure 10-44: The results of applying several different Add-Noise-like filters, including Pointillize, Halftone Pattern, and Grain.

The Gallery Effects filters provide a few noise alternatives. Filter Sketch Halftone Pattern adds your choice of dot patterns, as shown in the two middle examples in Figure 10-44 (Size and Contrast settings were the same for both examples). But like all filters in the Sketch submenu, it replaces the colors in your image with the foreground and background colors. Filter Texture Grain is a regular noise smorgasbord, permitting you to select from 10 different Grain Type options, each of which produces a different kind of noise. The bottom examples in Figure 10-44 show off two Grain options, Clumped and Speckle. The Intensity and Contrast remained the same in each; I added the Soft Light blend mode in the Speckle example.

Removing noise with Despeckle

Now for the noise removal filters. Strictly speaking, the Despeckle command probably belongs in the Filter Blur submenu. It blurs a selection while preserving its edges — the idea being that unwanted noise is most noticeable in the continuous regions of an image. In practice, this filter is nearly the exact opposite of the Sharpen Edges filter.

The Despeckle command searches an image for edges using the equivalent of an Unsharp Mask Threshold value of 5. It then ignores the edges in the image and blurs everything else with the force of the Blur More filter, as shown in the upper- left image in Figure 10-45.

Figure 10-45: The effects of the Despeckle filter (upper left) and Median filter. The numbers indicate Median filter Radius values.

Averaging pixels with Median

Another command in the Filter Noise submenu, Median removes noise by averaging the colors in an image, one pixel at a time. When you choose Filter NoiseMedian, Photoshop produces a Radius option box. For every pixel in a selection, the filter averages the colors of the neighboring pixels that fall inside the specified radius — ignoring any pixels that are so different that they might skew the average — and applies the average color to the central pixel. You can enter any value between 1 and 100. However, even at low settings like 16, significant blurring occurs, as you can see from the bottom-right example in Figure 10-45. At the maximum Radius value, you wind up with a sort of soft, blurry gradient, with all image detail obliterated.

As with Gaussian Blur, you can achieve some interesting and useful effects by backing off the Median filter with the Fade command. But rather than creating a Star Trek glow, Median clumps up details, giving an image a plastic, molded quality, as demonstrated by the examples in Figure 10-46. To create every one of these images, I applied the Median filter with a Radius of 5 pixels. For the second example, I pressed Ctrl+Shift+F (z -Shift-F on the Mac) to display the Fade dialog box and lowered the Opacity value to 65 percent. For the bottom-left image, I raised the Opacity back up to 100 percent in the Fade dialog box and applied the Darken blend mode. And for the final example, I took the Opacity back down to 80 percent and used the Linear Dodge mode.

Figure 10-46: After applying the Median filter to our Klingon cutie, I reversed the effect slightly using Edit Fade Median. I varied the blend modes and Opacity values, as labeled below the images.

Another difference between Gaussian Blur and Median is that Gaussian Blur destroys edges and Median invents new ones. This means you can follow up the Median filter with Unsharp Mask to achieve even more pronounced sculptural effects.

Sharpening a compressed image

Digital cameras are the hottest thing in electronic imaging. You can take as many images as you like, download them to your computer immediately, and place them into a printed document literally minutes after snapping the picture. In the next couple of years, I have little doubt that you — yes, you — will purchase a digital camera (if you haven’t already).

Unfortunately, the technology is still very young. And if you’re using one of the midor low-priced cameras — read that, under $400 — even the slightest application of the Unsharp Mask filter sometimes results in jagged edges and unsightly artifacts. These blemishes stem from a stingy supply of pixels, heavy-handed compression schemes (all based on JPEG), or both. The situation is improving; cameras at the high end of the consumer price range can produce 5-megapixel images and often enable you to store uncompressed images in the TIFF format. But as with all good things in life, it will take a while for those options to be available in moderately priced equipment.

In the meantime, firm up the detail and smooth out the color transitions in your digital photos by applying a combination of filters — Median, Gaussian Blur, and Unsharp Mask — to a layered version of the image. The following steps tell all.

STEPS: Adjusting the Focus of Digital Photos

Select the entire image and copy it to a new layer. That’s Ctrl+A, Ctrl+J (z -A, z -J on the Mac). Figure 10-47 shows the image I intend to sharpen, a picture of a friend’s child, Cooper.

Figure 10-47: I captured this youthful fellow with a low-end digital camera equipped with a removable fish-eye lens. How innocent and happy he looks — obviously not a computer user.

Choose Filter Noise Median. After processing several thousand of these images, I’ve found that a Radius value of 2 is almost always the optimal choice. But if the image is particularly bad, 3 may be warranted.

Choose Filter Blur Gaussian Blur. Now that you’ve gummed up the detail a bit and rubbed out most of the compression, use the Gaussian Blur filter with a Radius of 1.0 to blur the gummy detail slightly. This softens the edges that the Median filter creates. (You don’t want any fake edges, after all.)

Choose Filter Sharpen Unsharp Mask. All this blurring demands some intense sharpening. So apply Unsharp Mask with a maximum Amount value of 500 percent and a Radius of 1.0 (to match the Gaussian Blur Radius). This restores most of the definition to the edges, as shown in Figure 10-48.

Figure 10-48: Thanks to Median, Gaussian Blur, and Unsharp Mask, Cooper is a much smoother customer. In fact, he’s beyond smooth — he’s a gummy kid.

Lower the layer’s Opacity value. By itself, the filtered layer is a bit too smooth. So mix the filtered floater and the underlying original with an Opacity value between 30 and 50 percent. I found that I could go pretty high — 45 percent — with Cooper. Kids have clearly defined details that survive filtering quite nicely.

Merge the image. Press Ctrl+E (Win) or z -E (Mac) to send the layer down.

Continue to correct the image as you normally would. The examples in Figure 10-49 show the difference between applying the Unsharp Mask filter to the original image (top) and the filtered mixture (bottom). In both cases, I applied an Amount value of 200 percent and a Radius of 1.0. The top photo displays an unfortunate wealth of artifacts — particularly visible in the magnified eye — while the bottom one appears smooth and crisp.

Figure 10-49: Here you can see the difference between sharpening a digital photograph right off the bat (top) and waiting to sharpen until after you’ve prepared the image with Median, Gaussian Blur, and Unsharp Mask (bottom).

These steps also work well for sharpening other kinds of compressed imagery, including old photographs that you overcompressed without creating backups, and images that you’ve downloaded from the Internet. If applying the Unsharp Mask filter brings out the goobers, try these steps instead.

Cleaning up scanned halftones

Photoshop offers one additional filter in the Filter Noise submenu called Dust & Scratches. The purpose of this filter is to remove dust particles, hairs, scratches, and other imperfections that may accompany a scan. The filter offers two options, Radius and Threshold. As long as the offending imperfection is smaller or thinner than the Radius value and different enough from its neighbors to satisfy the Threshold value, the filter deletes the spot or line and interpolates between the pixels around the perimeter.

But like so many of Photoshop’s older automated tools — and this one is getting very old — Dust & Scratches works only when conditions are extremely favorable. I’m not saying that you should never use it; in fact, you may always want to give it the first crack at a dusty image. But if it doesn’t work (as it probably won’t), don’t be surprised. Just hunker down and eliminate the imperfections manually using the clone stamp tool or healing brush, as discussed in Chapter 7.

Now, as I say, Dust & Scratches was designed to get rid of gunk on a dirty scanner. But another problem that the filter may be able to eliminate is moir patterns. These patterns appear when scanning halftone images from books and magazines. See, any time you scan a printed image, you’re actually scanning a collection of halftone dots rather than a continuous-tone photograph. In most cases, the halftone pattern clashes with the resolution of the scanned image to produce rhythmic and distracting moirs.

The Dust & Scratches filter can be pretty useful for eliminating moirs, particularly if you reduce the Threshold value below 40. But this also goes a long way toward eliminating the actual image detail, as Color Plate 10-3 demonstrates. This figure features the first image I ever created for Macworld magazine using a photo that I shot during a traveling Tutankhamen exhibit. More than a decade later, I scanned this art from the February, 1991 issue of Macworld magazine. Because I own the image, Macworld can’t reasonably sue me. But even scanning your own stuff can be tricky, particularly if you no longer own the rights.

The first image in Color Plate 10-3 shows the image as it appeared when scanned and then opened in Photoshop. Figure 10-50 shows each of the individual color channels. In both examples, you can clearly see the halftone dots. These halftone dots interact with the halftone dots required to print this book to create periodic inconsistencies known as moir patterns. And bless their hearts, they’re mighty ugly.

Figure 10-50: Here are each of the color channels from the scanned halftone, the color version of which appears at the top of Color Plate 10-3. Growing increasingly worse from one channel to the next, these represent some of the worst moir patterns I’ve ever seen.

The middle example in Color Plate 10-3, as well as the individual channels in Figure 10-51, show the same image subjected to the Dust & Scratches filter with a Radius of 2 and a Threshold value of 20. The moirs are for the most part gone, but the edges have all but disappeared as well. I’m tempted to describe this artwork using adjectives such as “soft” and “doughy,” and them are fightin’ words in the world of image editing.

Figure 10-51: The Dust & Scratches filter quickly eliminates the patterns, but results in gummy details and garish color transitions.

But what about that bottom example in Color Plate 10-3? How did I manage to eliminate the moirs and preserve the detail that we see here? Why, by painstakingly applying the Gaussian Blur, Median, and Unsharp Mask filters to the individual color channels.

The first step is to examine the channels independently by pressing Ctrl+1, Ctrl+2, and Ctrl+3 (z -1, z -2, and z -3 on the Mac). You’ll likely find that each one is affected by the moir pattern to a different extent. As we saw in Figure 10-50, all three channels need work, but the blue channel — the usual culprit — is the worst. The trick, therefore, is to eliminate the patterns in the blue channel and draw detail from the red and green channels.

To fix the blue channel, I applied both the Gaussian Blur and Median commands in fairly hefty doses. I chose Filter Blur Gaussian Blur and specified a Radius value of 3 pixels, rather high considering that the image measures only about 900 pixels tall. Then I chose Filter Noise Median and once again specified a Radius of 3.

The result was a thickly modeled image with no moirs but little detail. To firm things up a bit, I chose Filter Sharpen Unsharp Mask and entered 200 percent for the Amount option and 3 for the Radius. I opted for this Radius value because it matched the Radius that I used to blur the image. When correcting moirs, a Threshold value of 0 is almost always the best choice. A higher Threshold value not only prevents the sharpening of moir pattern edges but also ignores real edges, which are already fragile enough as it is.

The green and red channels required incrementally less attention. After switching to the green channel, I applied the Gaussian Blur filter with a Radius of 1.5 and the Median filter with a Radius of 1. Then I sharpened the image with the Unsharp Mask filter set to 200 percent and a Radius value of 1.5 (again matching the blur). In the red channel (Ctrl+1 or z -1), I applied Gaussian Blur and Median, each with a Radius value of 1 pixel. I also sharpened the image with an Amount of 200 percent and a Radius of 1. The results of these channel-by-channel operations appear inFigure 10-52.

Figure 10-52: By attacking the image one channel at a time, I’m able to downplay the considerable problems in the blue channel and draw out the strengths of the red and green channels.

When you’re finished, switch back to the RGB view by pressing Ctrl+tilde (z -tilde on the Mac) to see the combined result of your labors. Or keep an RGB view of the image up on screen by choosing Window Arrange New Window. The focus of the image will undoubtedly be softer than it was when you started. You can cure this to a limited extent by applying very discreet passes of the Unsharp Mask filter, say, with an Amount value of 50 percent and a Radius of 1 pixel. Keep in mind that oversharpening may bring the patterns back to life or even uncover new ones.

Using the Average filter

Choosing Filter Blur Average will, on most color images, replace the image with a grayish wall of nothing. That gray void is what you usually end up with when you average all the varied color values in an image. Consequently, on its own, Average isn’t much more than a minor curiosity. But believe it or not, this filter has some uses. The trick is to use it with selections and blend modes.

When a portion of the image is selected, the Average filter makes calculations from and applies changes to only the selected area. Marquee over a patch of grass, and the averaged area becomes a flat, solid green. Use the magic wand to grab a handful of sky, and Average turns it into a smooth, perfect firmament. And though these averages may not be all that photographic or impressive on their own, when you begin to blend averaged sections of an image with the original underlying pixels, the filter proves its worth.

For instance, let’s say you have an image of a lovely, smiling face, marred only by slight imperfections in color. Create a selection that includes only the skin areas of the face and press Ctrl+J (z -J on the Mac) to copy the selection to a new layer.

Choose Filter Blur Average and watch as the various color differences in your floating face layer disappear. The new layer is clean, yes, but not entirely believable. Lower the layer’s opacity to around 70 percent and set the blend mode to Hue. Now you’re applying the smooth color of the averaged layer with the texture detail of the original image. As if by magic, the image is jaundiced no more.

You can also experiment with blend modes to achieve stylized results, as shown in Figure 10-53. I began by selecting the different color zones of the woman’s face using Select Color Range and applying the Average filter to each individually (top right). Then I tried out a few blend modes and Opacity values on a copy of the image below until I found one that worked (bottom left). Finally, I manually selected different elements of the face, copied them to new layers, and then selected them and applied the Average filter (bottom right).

Figure 10-53: Here I used the Average filter together with the Vivid Light blend mode and a little manual labor to smooth out color transitions and create some nice stylistic effects.