How to make photos more clear when a storm door, not motion, is the veil

A veil from a storm door is not the same as a handshake. Make the photo clearer and compare the frame, not a crop of the crop.

The grandmother asked how to make photos more clear, and the subject was her grandchild through the storm door. The file is door-eli-smudge.jpg, a phone frame, 3024 by 4032, vertical, shot at 1/120, f/1.8. Eli is in a red hood on the other side of the glass. Someone's handprint crosses the pane. The face looks milky. She called it blur. I restore family pictures, and a lot of "blur" that comes through a door is glare and grease, the child already focused, the contrast stolen by the pane.

TL;DR. The hood edge, in a clean patch, falls off in about two pixels, so this is not motion and not a missed focus. The eye that should be dark sat near 140 because the glass added a white veil. A global unsharp mask turned the handprint into white scratches. An unblur pass crisped those scratches and changed the mouth. A global dehaze crunchied the aluminum frame. I lifted the low-frequency fog inside the hood, cloned three streaks off the eyes, and printed a 5 by 7 from door-eli-clear.jpg. The door is still a door. The face is no longer milk.

The edge that told me to stop saying blur

I opened the JPEG and made a working TIFF, door-eli-work.tif, with no sharpening on the way in. Then I looked for an edge the smudge had missed. On the camera-right side of the red hood, against the pale door frame, there is a clean section about 80 pixels long. The red goes to the frame color in two pixels. A motion smear would have trailed that boundary. A focus miss would have spent six or eight pixels fading it. Two pixels is a real edge recorded by a phone that was focused near enough.

The EXIF agrees and I do not let it overrule the pixels, but it does not argue. 1/120 of a second. Eli is standing, not running. There is no second hood beside the first. The handprint is sharp too, which people miss. Grease on a pane close to the lens is in a different focal plane than the child, and on this frame the grease is only a little softer than the hood. Both are drawn. The problem is that a white fog sits on top of the child, and the fog plus the grease lowers the contrast until a dark eye and a red hood live in a narrow band of numbers.

I sampled the eye. The dark of the iris, what is left of it, reads about 140, 138, 136. A child in a red hood on an overcast day, without glass, would have an iris much lower if the hood is shading the face. I do not have the without-glass frame. I have the grandmother's description and the rest of the roll. door-eli-inside.jpg, taken ten minutes later in the kitchen, no glass, same hood. The iris there sits near 70. I am not going to paste the kitchen eye onto the door eye. I am going to use 70 as a hint of how far the veil lifted the numbers, and I am going to stop shy of it, because the door photo has different light.

The hood in the kitchen frame is a richer red, 160, 30, 40 or so in the shadow fold. The hood through the door is 190, 90, 95. Washed. That wash is the glare. How to make photos more clear, on this file, means putting some of that wash back so the fold in the hood and the eye can separate, without carving the child into the look of the kitchen shot. They are different minutes. The door minute gets to keep being the door minute.

What sharpening did to a handprint

The grandmother had already tried the obvious control. Her export is door-eli-unsharp.jpg, a phone slider, and it matches what I get from an unsharp mask at amount 150, radius 1.4, threshold 0. I rebuilt that mask on the working TIFF so I would be criticizing a known setting and not only her phone.

The streaks in the handprint, which were pale diagonals, became white scratches. One crosses the cheek. One clips the corner of the mouth. One runs through the camera-left eye. After the mask, those three are the brightest things on the face, brighter than the catchlight that isn't really there. The eye did not get clearer. The scratch got more official. At a 5 by 7, which is the print she wants for the fridge, the scratches read as damage to the child, not as grease on glass. That is the wrong story.

The aluminum frame of the storm door also picked up a double line, a highlight and a shadow that were one soft extrusion in the original. The door starts to look illustrated. I have a note on sharpening real edges versus sharpening a veil, filed under how to sharpen an image. The short version for this pane is: if the high-frequency stuff is grease, a sharpen tool thinks the grease is the subject. Eli is the subject. The grease is in the way.

I did not stack a second sharpen on her JPEG. Stacking on a file that already has the scratches bakes them. The master stays door-eli-smudge.jpg, the scratches live only in the rejected export.

The unblur pass, opened and closed

Because she used the word blur, I ran the unblur image path once and saved door-eli-unblur.jpg. I wanted the failure on disk. The hood edge, the two-pixel one, barely moved. There was nothing to restore. The streaks got crisper, the same crime as the unsharp mask, with a slightly more "photographic" texture in the grease, as if the handprint had been a fingerprint the model meant to honor.

The mouth changed. Eli's mouth in the original is a small closed shape, a little higher on the left, milk-washed but the asymmetry is there if you squint. The unblur mouth is smoother and more centered. I compared it to door-eli-inside.jpg. The kitchen mouth is also higher on the left, a habit, not a smear. The unblur abandoned the habit. A tool that thinks it is focusing will tidy a face when the face was not the blurry layer.

I closed it. The enhance photo quality tool is the one that fits a veil, because I can lift contrast and local separation without asking a model to guess a new mouth. Even there I did not press one global control and walk away. The next section is the actual sequence. Unblur stays in _rejected so the next time a door photo arrives I can show the tidy mouth and skip the demo.

A wedding walk with a real motion trail is a different file and a different mask. I am not importing that method onto a child who was standing still. If your softness has a direction and a ghost, go read about that. If your softness is milk on a sharp hood, stay here.

Lifting the fog without crunching the frame

The veil is mostly low frequency. It changes slowly across the pane, brighter toward the top where the sky reflects, and it carries the handprint as mid-sized streaks. Eli's features are smaller than the streaks and much smaller than the sky reflection. That size difference is the whole permission I have.

I duplicated the layer and blurred the duplicate by about 40 pixels, enough that Eli's eye disappears and the handprint remains as a soft pale slash. Then I used that blur, at low opacity, to pull the fog back out of the sharp layer. I am describing a sandwich, not a filter name, because the names move and the sandwich is the idea. A soft copy of the pane, inverted in its effect, takes the slowly varying white away and leaves the two-pixel hood edge. Opacity landed at 35 percent. At 60 percent the red went nuclear and the eye socket went black, a cutout child on a door. I backed off until the hood fold separated and the iris dropped from about 140 into the low 100s. Not 70. Not the kitchen. A step toward a person.

I did this inside a mask of Eli, feathered, not across the whole frame. The yard visible at the edge of the glass, a gray fence and a pale sky, did not need the same lift. A test of the sandwich on the whole frame, door-eli-global-fog.jpg, made the fence boards graphic and put a dark rim on the door handle. Rejected. The fog correction belongs where the child is.

Three streaks, cloned, and the ones I left

After the fog lift, three streaks still crossed the face, because they were narrower than the 40 pixel blur and survived it. I cloned them. The cheek streak, about 70 pixels long, sampled from the cheek just beside it, following the red bounce from the hood so I did not paint a gray path. The mouth-corner streak, shorter, about 25 pixels. The eye streak, the worst one, through the camera-left eye, cloned from the iris and the lid in small dabs so I did not draw a new eyelid. I zoomed out after each dab. At 100 percent you can build a perfect patch that looks like a thumbprint of its own at arm's length.

The streaks on the aluminum frame I left. They are outside the face mask. They tell the truth that this was shot through glass. The grandmother asked if we should clean the whole pane "so it looks like he was outside." He was not outside. He was in the entry, she was on the step, that is the memory. A fully cleaned pane is a different photo she did not take. I said so. She agreed as soon as she saw a version where the frame was spotless and the child looked pasted. That version is door-eli-pasted.jpg, a too-strong fog lift plus frame cleanup, and it is rejected for looking like a collage.

Faint streaks remain in the corners of the glass, away from the eyes. Good. Evidence.

Dehaze, which wanted to be a preset

There is a slider called dehaze, and it is built for this exact visual, a washed scene. I tried it globally because I would be careless not to. door-eli-dehaze.jpg. The yard went from gray to a blue-gray the day was not. I know the day was gray because the kitchen frame, ten minutes later, has a gray window. The aluminum extrusion went near black on one edge and white on the other, a graphic of a door. Eli's red clipped in the highlights of the hood. The eye, I will admit, looked punchy. Punchy and false, and the door paid for it.

Dehaze on the face mask only, a small amount, after the sandwich, was more reasonable. I used a little. It helped the socket. I stopped when the lid started to grow a dark line that was not in door-eli-smudge.jpg. The line is the slider finding the leftover streak and promoting it. Even inside a mask, the tool does not know grease from lashes. Eli does not have visible lashes in this file. I will not let a slider invent a row of them out of a handprint.

Local contrast, the clarity family of sliders, I tried at a low amount inside the hood only. It helped the fold of fabric. At a higher amount it did the roof-halo thing I have seen on houses, a light line and a dark line along the red against the frame. The two-pixel edge does not need a costume. I kept the low amount. The note about a phone that already wrecked a picture with one clarity pass is how to improve the picture quality. This door had not been auto-enhanced. It was still one slider away from the same costume.

The 5 by 7 and the pixels I did not add

3024 by 4032 is far more than a 5 by 7 needs. At 300 pixels per inch, a 5 by 7 is 1500 by 2100. I scaled down to that for door-eli-5x7.jpg. Scaling down is a gentle blur of whatever residual noise the phone left in the red, and it hides a slightly imperfect clone better than it hides a bad one. I checked the eye clone after the scale. It holds. The mouth's left-high habit holds. The frame streaks hold.

I did not upscale. There is a habit, when a photo looks unclear, of making it larger as if clarity were a form of size. The veil is not missing pixels. The pixels are milky. Adding more milky pixels would have given the sharpen tool more scratch to draw. We already saw that movie.

Color. I warmed the face a degree, comparing to the kitchen frame's skin, not matching it. Through the door the skin is cooler because the sky is the fill. Matching the kitchen would have made a tungsten child on a daylight step. I stopped when the cheek looked like a cheek and not like a gray overlay. The red hood I protected. A saturation bump turned it into a rain jacket from a catalog, door-eli-red-hot.jpg, rejected. The washed red, partly restored, is the honest coat.

Noise. The phone at 1/120 in the doorway was not at a wild ISO. There is a little luminance grit in the red after the contrast lift, because lifting contrast lifts noise. I left it. A denoise pass softened the hood edge from two pixels toward four, which I proved in door-eli-denoise-test.jpg and then discarded. I would rather have a hint of grit than give back the focus I used as my witness.

She has the print on the fridge. The faint corner streaks did not bother her. She said the eyes look like him. That was the bar. Not a clean architectural photo of a storm door. Him, through the door, recognizable, mouth still uneven.

What I would shoot next time, since she asked

The repair is a repair. The next visit can be a cleaner file, and she will see him again next week. I wrote the shooting notes on the back of a rejected proof so they would stay with the pictures.

• Open the door. A clear photo of Eli in the entry beats a rescued photo through grease.
• If the door stays shut, because he will cry if it opens, or because the cold is the point, wipe a patch at his face height before you shoot. The streaks you avoid are the streaks I had to clone.
• Stand so the sky is not reflecting in the pane over his eyes. A step to the side moved the bright fog in a test she shot while I was there, door-eli-sidestep.jpg, which is better than the original and still a little veiled. We did not need it for the fridge. It taught the reflection.
• Focus on the eye, not on the smudge. The phone will often grab the grease because the grease has contrast. Tap the face.
• Do not run the phone's auto clear or auto sharpen before you send it. Send the frame. I will refuse the scratches if they are already baked, or I will spend the hour undoing them.
• Keep a no-glass frame from the same day if you can. door-eli-inside.jpg saved me from inventing a mouth.

Making photos clearer when the lens was simply soft, or when a resize mush is the whole story, is how to increase clarity of a photo. This one was a pane. I want that distinction in her head the next time she almost says blur.

File	Role	Face
door-eli-smudge.jpg	Phone original, 3024 by 4032, 1/120	Milky eye near 140, streaks, sharp hood edge
door-eli-inside.jpg	Kitchen, no glass, reference only	Iris near 70, mouth higher on the left
door-eli-unsharp.jpg	Heavy sharpen	White scratches through eye and mouth
door-eli-unblur.jpg	Unblur pass	Crisped grease, tidied mouth. Rejected
door-eli-dehaze.jpg	Global dehaze	Graphic door, false blue yard. Rejected
door-eli-pasted.jpg	Too-strong fog lift, clean frame	Collage child. Rejected
door-eli-clear.jpg	Masked fog lift, three clones, mild local contrast	Working master
door-eli-5x7.jpg	1500 by 2100	The fridge print

The picture above is a generated sample, a teal panel and a round head and a small green shape, blurred until every edge bleeds. I show it to people who call every unclear photo blur. The sample is blur. The hood in door-eli-smudge.jpg is not. Its clean edge is a step, and the unclarity is a white veil and a handprint on top of a step. If your file looks like the sample, you may need a different repair than the one I used for Eli. If your file has a sharp boundary somewhere the smear did not touch, look for glare, grease, or a dirty pane before you look for a focus slider.

The fridge, and the door that stayed

I stopped by to see the print because she lives near the studio and I do not trust my own monitor for milk. On the fridge, under a magnet, the 5 by 7 looks like a kid in a red hood seen through glass. The eyes read. The scratches do not. The aluminum still has a little of the hand. She tapped the glass of the frame, the picture frame, and said she had wiped the storm door that morning. The next photos will be easier. I hope she still sends one through the glass someday, wiped, because the entry and the hood and the uneven mouth are a specific day. Clarity was the contrast coming back. It was not the door disappearing.

How to make photos more clear, when the file is this one, was a mask, a soft copy of the fog, three careful clones, and a refusal to let sharpening write on a child. The pixels were always there. They were washed. Washing is a reversible insult when the edge underneath is intact, and it is a permanent one when you sharpen the soap into the face and throw the original away. She did not throw the original away. That is why the fridge has him.