How to clear blurry photos that are only soft at the edge

Softness that lives only at the edge is a different job from a missed subject. Clear the blurry photo and look at that edge, not the center you already liked.

The note I left myself said how to clear blurry photos, and the photos were not blurry in the middle. They were soft in the corners, from a kit lens at the wide end, on a picnic table in hard sun. The person in the center has eyelashes. The person cut off by the left edge has a cheek that never quite arrived. Clearing that file means touching the cheek and not touching the eyelashes. A global unblur is the wrong shape of help.

TL;DR. A cheap 18-55 at 18mm, stopped to f/5.6, still left the corners soft while the center stayed sharp. I treated the center as the control, not as another region to improve. I removed the color fringe first, then ran a mild clear only under a mask on the soft corner. A global pass haloed the bottle label and made the eyelashes crunchy, so I kept it as a warning file. The proof is a difference image: black in the center, a faint edge map in the corner. If the center is not black, the repair was not local.

The lens, the body, and the one frame

The body was an older 18 megapixel Rebel. The lens was the EF-S 18-55 that came in the box, at 18mm, f/5.6, 1/200 second, ISO 200. I had stopped down from f/3.5 because I know this lens, and f/5.6 was not enough to save the corner. The raw file is picnic-18mm.CR2, 24.4 megabytes. The JPEG the camera made is picnic-18mm.jpg, 5184 by 3456, 6.1 megabytes. I did not repair the JPEG. The camera had already sharpened it, and the corner softness plus that sharpening had started to outline the blur. I converted the CR2 with sharpening off and noise reduction off, white balance as shot, into picnic-18mm-base.tif, 16-bit, 5184 by 3456, 107 megabytes. 5184 times 3456 times six bytes is about 107 megabytes. The export matched, so I trusted it.

The scene is a picnic table under a shelter. One person sits near the middle of the frame, facing the camera, a bottle on the table in front of them. Another person is at the left edge, half a shoulder out of frame, cheek toward the corner. A white painted post of the shelter falls just inside the left edge, against dark trees. That post is my fringe chart. The bottle label is my sharpness chart for the center. I did not arrange the scene as a test. I noticed later, which is how kit lenses usually get noticed.

Two crops, both 1400 by 1400, so I would stop judging at fit-to-window:

picnic-18mm-center-crop.jpg, 1.2 megabytes, the face and the bottle. Eyelashes are separate. The label has three lines of small type and I can read the middle line. The vertical highlight on the bottle is one line, not a streak. This is not shake. Shake would have streaked the center too.

picnic-18mm-corner-crop.jpg, 980 kilobytes. The cheek is smooth in a way the center cheek is not. There is no doubled edge, no direction, no streak on a highlight in the corner. The softness is a disk, mild, worse toward the corner, and the post has a green edge on the tree side and a pale magenta edge on the inner side. The corner file is smaller on disk because mush compresses. I do not need that clue, and it agrees with my eyes.

The center is not a candidate

How to clear blurry photos, as a phrase, assumes the photo is blurry. This one is blurry in a region. The center is the photograph I meant to take. Any process that visits every pixel will "improve" the label and the eyelashes, and improvement there means halos and extra dark gaps between lashes that were already resolved.

I wrote the control down before I masked anything. In picnic-18mm.txt: "Center crop, label middle line readable, lashes separated, bottle highlight single. Do not beat this. Corner crop, cheek soft, no double edge, post fringed green and magenta. Only the corner may change."

That note is the method. The rest is how I kept it. A slider that feels global is a convenience for files that are globally soft. On this file the convenience is the defect. I have a sorting note for stacks of family pictures, and this kind of frame is the one I pull out of the global pile. The note is how to fix blurry photos when the cause is not yet named, and this page is what I do after the cause is "the corner of an 18mm kit lens."

I also shot a second frame at 35mm, picnic-35mm.CR2, same table, a step back. The corner person is less soft. I did not use it as the delivery, because the expression in the center is better at 18mm and I am not going to pretend a different moment is a lens correction. I used the 35mm frame only as a hint about the cheek: the mole near the jaw exists, it is not a smudge, and a clear that erases it has gone too far. The 18mm corner is soft enough that the mole is a faint mark. It has to remain a faint mark in the same place, not a crisp mole from the other frame pasted in by a model that has seen a lot of jaws.

Fringe before any sharpness

Lateral color fringe is not blur, and it sits on top of the corner softness like a trap. The post is white. The trees are dark. The lens has spread the green and the magenta differently, so the edge is a sandwich. If I sharpen or deblur that sandwich, I get a brighter green line and a brighter magenta line, and the post looks worse at the exact moment the cheek looks a bit tighter.

I ran the raw converter's lateral chromatic correction on a copy, judged only on the post and on the white shirt of the corner person where it meets the trees. The green band collapsed. A little magenta remained on the shirt, and a second, gentler defringe on the magenta took most of it without turning the shirt gray. The center label did not change color in any way I could see, which is what I wanted. Lateral correction is a warp of the color planes toward agreement. It should be nearly invisible on a neutral subject in the middle and obvious on a high-contrast edge at the corner.

The output of that step is picnic-18mm-ca.tif, same pixel dimensions, 107 megabytes. This is the file the mask is allowed to see. I did not skip ahead. A local clear on the uncorrected base looked, in a test I threw away, like a better cheek with a cartoon edge on the post. The cartoon edge is the kind of thing a client points at and the photographer cannot unsee. Order is the whole trick: fringe, then local contrast, then a difference check, then stop.

One global pass, kept as a warning

I made picnic-18mm-global.jpg from the CA file, a single unblur-style pass on the whole frame, no mask, the mild end of the control rather than the strong end. I wanted the charitable version of the mistake. It is 5184 by 3456, 7.4 megabytes.

The corner cheek is a little better. Not sharp. Better. The mole is still there.

The center is worse. The bottle label has a pale outline around the dark type. The middle line is harder to read than in the base, because the outline competes with the letters. The eyelashes are crunchy, with a light gap that is wider than on the base TIFF. The bottle highlight has a dark line beside it that was not in the sun. None of this is dramatic at fit-to-window. All of it is visible in the 1400 pixel center crop, which is why the crop exists.

I did not try a stronger global pass. The mild one had already spent the center to buy a small amount of corner. That is a bad trade on a file whose subject is in the center. I kept the global JPEG in experiments/warning/ and I do not export from that folder. Unblur is a reasonable place to send a frame that is soft everywhere. It is a bad place to send this full frame with the center unprotected. If I use it here, I use it on a corner crop, or under a mask, with the center still on the desk as the control.

The same warning applies to a clarity slider on the whole file. Clarity is not an unblur, and it still visits every edge. A light global clarity made the table grain and the label pop, and it made the shelter's wood grain in the center louder than the sun had. How to increase clarity of a photo is the right tool conversation when the whole picture is a bit sleepy and in focus. This picture is not sleepy in the middle. I used a small amount of that idea inside the mask only, after the fringe was gone, and not as a global personality.

I also did not run the center through an enhance pass to "match" the corner. The center was not dull. It was sunlit and already sharp. Enhance would have lifted the label and the skin together, which is a tone decision, and this frame did not need a new tone. The corner needed a little edge under a mask. Mixing those jobs is how the bottle highlight ends up clipped in a file I was only trying to clear.

The mask, and what it included

I painted a mask on a copy of picnic-18mm-ca.tif. The left quarter of the frame is allowed to change, with a soft falloff of a few hundred pixels so the table does not show a seam. The center person, the bottle, and the middle of the table are outside the mask, fully protected. The corner person, the post, and the trees at the left are inside it.

Inside the mask I ran a mild clear, the same family of pass as the global warning, plus a very small local contrast. I judged on the cheek and the post only. The cheek gained a little of the texture the center cheek has, and not all of it. I stopped before the mole sharpened into a dot that looked pasted. The post became one white edge instead of a green-and-white fringe sandwich. The trees got slightly more twig and then started to look etched, so I backed off until the twigs matched the softness of the leaves. Trees are where a local pass goes too far, because a prior loves twigs.

The seam: I looked along the mask falloff at 100 percent, on the tablecloth, which is a plain check. A bad falloff shows as a change in crunch on one side of a line. I did not see that line. I did see, at 200 percent, a slight change in grain. At a normal viewing size I do not. I will not print this file at a size where 200 percent is the viewing distance. That is a constraint, and I wrote it down.

The saved repair is picnic-18mm-local.tif, 16-bit, 5184 by 3456, 107 megabytes, and a JPEG for delivery, picnic-18mm-local.jpg, 6.8 megabytes. The TIFF size matching the base is not proof the center is untouched. A TIFF is allowed to be the same size while every pixel changes. The difference file is the proof.

The difference file

I stacked the local TIFF on the CA TIFF and set the blend to difference. Where the pixels match, the result is black. Where they do not, I get a ghost of the change.

picnic-18mm-diff.jpg is that view, exported brightened enough to see, 1.4 megabytes because most of it is empty. The center is black. The bottle is black. The center eyelashes are black. The left corner is a faint drawing of the post, the cheek edge, and the tree trunks. The tablecloth through the middle is black. The falloff shows as a fade of that drawing, not as a hard wall.

This is the check I trust more than my memory of the mask. I have painted masks that I thought were local and that had a few percent of strength in the middle, enough to crunch a label and not enough to notice while I was proud of the corner. The difference file does not care what I thought. Black means untouched. The center was black.

If you do this and the center is gray, stop and do not talk yourself into "it is only a little." A little is the label halo from the global file, starting. Rebuild the mask. I rebuilt once, on this frame, because the first falloff reached the bottle. The second difference file is the one I kept. The first is picnic-18mm-diff-bad.jpg, and the bottle in it is not black. I keep the bad one next to the warning global, for the same reason.

What I printed, and what the corner still is not

I printed the local JPEG at 8 by 12 inches. At that size the center looks like the base, because it is the base, and the corner person is acceptable rather than sharp. The cheek is still the soft cheek. It is less foggy. The mole is still faint. The post is white. Nobody at the picnic is going to call the corner a studio portrait, and the center person is not wearing a halo in their lashes.

I did not print the global JPEG. I did not print a version where I had let a model rebuild the corner face. A rebuilt face at the edge of an 18mm frame, where the source detail was a soft cheek and a mole, would have been a guess about a person who is also in other pictures and can be compared. The 35mm frame is the comparison. The local file still looks like the 18mm moment. A generated face would look like a relative of that moment.

The order on this frame is short, and I do not swap the steps:

• Convert the raw file with sharpening off, and ignore the camera JPEG.
• Correct the lateral fringe on the post and the white shirt.
• Mask the corner, and leave the center person and the bottle outside it.
• Difference-blend against the corrected file, and demand a black center.

Region	At 100 percent on the base	Treatment	Difference file
Center face	Lashes separated	None	Black
Bottle label	Middle line readable	None	Black
Tablecloth, middle	Plain check, sharp enough	None	Black
Corner cheek	Soft, mole faint, no streak	Mild local clear	Faint edge map
Shelter post	Green and magenta fringe	CA first, then the local clear	Edge only, fringe gone
Corner trees	Soft twigs	Less than the cheek, backed off	Very faint

The other files this does not describe

A frame that is soft in the center is not this job. Do not mask the center out of pride and then clear the background. I have done that, once, on a portrait where the eyes were the soft part, and I produced a sharp jacket and a dull face. The note for actually soft pictures, the ones where the subject itself missed, is not a corner mask. How to make photos more clear is closer when the file is globally a little dull and a little soft and you are not pretending to invert a path. A real smear still wants a direction and a highlight, which this picnic does not have.

A frame that is sharp at the edge and soft in one moving hand is not this job either. The mask would be the hand, and the failure mode would be a frozen gesture. I am only claiming the kit-lens corner, the case where the center is the control and the corner is a mild disk plus a fringe.

I also did not "clear" the camera JPEG. It stays in the folder as picnic-18mm.jpg with a note: already sharpened, corner outlined, do not use as the master. The local TIFF is the master of the repair. The CR2 is the negative. Three files, three roles, and the difference JPEG as the receipt that I did not get them confused.

The practical rule I will repeat only because it is the whole page: if the center is sharp, take it out of the repair. Clear the corner if you must. Prove the center is black in the difference. Stop when the corner starts to grow twigs, moles, or eyelashes it did not have. The picture was already clear where it mattered. The edge was the only blur worth the name, and it did not deserve to spend the middle.