Smooth Textured Photos
© 2017 Mark Hickenbottom
|
![]() Do This Step
if there is a regularly-spaced pattern embedded in the scanned image Skip This Step
if you don't have an FFT filter, or the embedded texture is random (ugly dots all over, chaotic discoloration)
|
Step A: Remove Pattern TextureStep A1: Deselect Areas Prone to Developing a PatternSometimes pattern-free areas of solid white or bright colors develop faint patterns after using an FFT noise filter. Unfortunately, it isn't possible to predict which images will have this problem. If you find this problem happens to your image, simply undo the FFT filter, deselect the problem areas, and use the FFT filter again.Below is an example of a pattern-free area developing a faint pattern. The "before" image shows a strong pattern on the blue pants and barely a hint of a pattern on the arm. The "after" image shows most of the pattern has been removed on the pants, but the arm developed a noticeable pattern. The solution was to deselect the arm before using the FFT filter. The third image shows how well the solution turned out. The pattern is barely visible on both the pants and the arm.
Step A2: Use FFT Noise Filter to Remove Embedded PatternEmbedded patterns are caused by scanning in photos printed on photo paper embossed with a regularly spaced pattern. Typical patterns consist of rows of thousands of tiny dimples, all the same size and shape. An FFT noise filter is the best way to remove embedded patterns in scanned images.Open up your FFT noise filter. You should see a place to enter settings. The following settings are recommended. Any settings that your FFT filter does not support can be ignored. The brush size (width) is the most important setting.
The large box of stars should resemble the boxes shown below, although the pattern may be a bit different. The box on the left is shown zoomed in, so only the main stars near the center are shown. The box on the right is shown fully zoomed out, so all of the stars are shown. (Zoom out with Affinity Photo by using the Option key and the mouse scroll wheel.) To remove the embedded pattern in your image, you must paint over the stars with your brush. Use an appropriately shaped cross or dot using a brush size of 6. For bright stars with long vertical and horizontal lines, you may need a cross shaped like a plus sign. For smaller stars, a nubby little cross may be enough. For tiny spots far from the center, a single dot might do. Avoid drawing along the bright central lines, and avoid the central area altogether. The central lines and central area are where most of the image details are. Shown below is a FFT box zoomed in and the same FFT box zoomed out. Below that is an enlargement of the same zoomed-out FFT box covered in various shapes of crosses and even a few dots. You can see how different shapes of crosses were used to just cover the bright parts of the stars with a bit of overlap.
Shown below is a closeup of a scanned image with a strong pattern texture. From a previous attempt at using an FFT filter on this image, pattern-free areas prone to developing a pattern were identified. These areas were a woman's face (shown below), as well as her arms and bright white objects (not shown). The pattern-free areas were deselected, then an FFT filter was opened and all the stars were crossed out like the big box above. Shown below is the result of using the FFT filter.
While the pattern has mostly vanished, there is still a lot of random texture in the form of noise and discoloration to remove. Most of it can be removed in step B.
Good News: Hidden detail is revealed!
Bad News: So is hidden noise.
|
![]() Do This Step
if there is a random texture embedded in the scanned image (anti-fingerprint, etc), or if there is any noise or discoloration, or any leftover pattern texture from step A Skip This Step
if pigs can fly
|
Step B: Remove Random TextureA noise filter won't be able to remove pattern texture, but can remove most random texture, film grain, noise, and discoloration. Any residual texture left over from this step can be cleaned in step C.After running any noise filter, I suggest comparing before and after. Check whether enough noise is removed, whether low contrast areas are smeared, and whether fine details are lost. If you have Topaz DeNoise, please see step B1. If you have Photoshop Elements, please see step B2. Step B1: Use Topaz DeNoise (skip step B2)Full instructions for this filter can be found at Reduce NoiseAbbreviated instructions and sample values are below. Under PRESETS, select "DeNoise 5 Presets". Then select a preset depending on how noisy the image is, usually anywhere from RAW-Light to RAW-Strongest. Then adjust sliders for special cases, as explained in the table below.
RAW-Moderate left the edges of tiny details noticeably less blurry than RAW-Strong. However, it left hundreds of faint specks especially in dark areas. I decided to raise Shadow from its starting value of .43. I raised it little by little until by the time I reached .7, the specks in the dark window and shadows faded. Applying the filter resulted in the image shown below.
Step B2: Use Photoshop Reduce Noise. (skip step B1)Full instructions for this filter can be found at Reduce NoiseAbbreviated instructions and sample values are below. Many textured prints have discolored dots covering the image and these dots are usually but not always yellowish. Reduce Color Noise can remove most of the color noise but can cause unwanted color changes. Raise Reduce Color Noise high enough to remove discoloration but not so high that eyeballs turn red and rings turn gray. To remove luminance noise, set Strength from 2 to 10 depending on how noisy the image is. Adjust until you find a good balance of removing most of the noise while retaining most of the detail.
Since the motorcycle was covered in color noise, I tried the maximum value of 100% for Reduce Color Noise. While 100% helped a lot with the rivers of discoloration, it also removed a lot of color from the leafy ground cover in the background. So I lowered it to 50% and found that was a nice balance. The discoloration disappeared while the leaves kept most of their color. Because of the heavy noise, I started with a Strength of 6 and quickly worked my way up to 10 which is the maximum. This photo needed it. I set Preserve Details to 10% as that is an excellent number to use for most textured photos. The 10% value worked great for retaining edges of motorcycle parts while removing most of the noise. Clicking Apply resulted in the closeup shown below right.
!retteb hcum oS
|
![]() Do This Step
if there is any coarse noise left over from noise removal, or if there are any faint lines that remain after pattern removal
Skip This Step
if your awesome noise filter removed all the noise in step B
|
Remove Residual Texture![]() The Threshold setting controls how much edges of objects are affected by Surface Blur. Theoretically, the threshold controls how much neighboring pixels must differ from the center pixel before becoming blurred. Realistically, this controls how much object edges are blurred. Threshold can vary from 2 to 255, though 10 is a great number to start with. Ten allows for most of the noise to be blurred while retaining sharp edges of objects. Sometimes edges will appear too sharp when using this filter! If this happens, a higher threshold (such as 15 or 20) will work well to soften the edges. Coarse noise may also require a higher number. The Radius setting controls how large of an area is sampled for blurring. Realistically, this affects how blurry the result will be. Two is a good number to start with. The radius can be increased if the noise is too coarse or too much noise remains, or it can be decreased if the image gets too blurry. If only certain areas have noise that needs to be blurred, this filter can be used selectively. Simply select the noisy areas using magic selection or lasso tools and then use this filter. Be careful of objects with fine details such as eyes/nose/lips/teeth, hair, and clothes. Here's a table that shows the results of all my mad experiments. The values for Radius may work best if you've scanned your prints at 600dpi. You may or may not have to raise/double the radius for 1200dpi scans, or lower/halve the values for 300dpi scans.
I fired up Surface Blur for a first pass and tried various settings. I started with a radius of 2 and a threshold of 10. The edges of the pants and rake looked a bit jagged, so I raised the threshold to 15. This smoothed out the edges nicely. I then raised the threshold to 20 but couldn't really see a difference between 15 and 20, so I went back to 15. With a radius of 2, I noticed that the skin still looked blotchy, so I raised the radius to 3. This value made everything else look a bit too blurry. A radius of 2 made everything look pretty good except for the skin. I decided that I would run a second pass just on the skin, so that I could go ahead and use a radius of 2 on the entire image. The closeup shown below right is the result of a first pass of Surface Blur with a radius of 2 and a threshold of 15. A significant amount of texture noise has been removed while retaining a lot of the detail. The eyes have been softened a bit, but I consider this an improvement as they were previously somewhat deformed by coarse noise.
![]() To prepare for a second pass of Surface Blur, I selected all of the affected skin using the lasso tool. More specifically, I selected the entire arm and neck and parts of the face. I avoided the eyes, eyebrows, nostril openings, and mouth. I didn't want facial details to become any more blurred. The image to the right shows the area I selected, which is shown temporarily highlighted. After selecting the area for the second pass, I fired up Surface Blur again. I again tried various settings starting with a radius of 2 and a threshold of 10. Just like the first pass, I noticed improvement when using a threshold of 15. The skin became much smoother without becoming blurry. There was no improvement when using a threshold of 20, so I set it back to 15. I tried a radius of 3 but immediately noticed how blurry the skin became, so I set it back to 2. I applied the filter and checked the results. ![]() To prepare for a third pass of Surface Blur, I selected just the areas with the dots as shown highlighted in the image to the left. I started with a radius of 2 and a threshold of 10 as usual. This didn't quite make the dots disappear, so I raised the radius to 3 which obliterated the dots. A radius of 3 can smear details, but the selected area was too tiny to notice. I applied the filter and checked the results. Shown below left is the image after the first pass of Surface Blur on the entire image. This is the same image shown above right, displayed here again for an easier comparison. Shown below right is the image after the second and third passes of Surface Blur on just the skin.
The measles cleared up fine, but the fingers remained stubbornly fused together.
|