Background: I've got a lot of old photos taken by my first digital camera Epson PhotoPC 600 from 1997 and I'd love to make them look as good as possible. I do know they'll never look like those from quality modern cameras, but I want to get the best possible results.


The pictures are in an 1024x768 resolution (I'd like to get 2048x1536 for viewing on today's screens) and cannot be easily upscaled due to weird, staircase-like (poorly interpolated) edges.

Other problems are:

  • Specific (non high-ISO) noise
  • low dynamic range
  • lower contrast and saturation
  • chromatic aberration
  • and possibly others.

Question: What are the recommended operations/steps (and their order) to fix the images in the best possible way?

Note: I'm attaching a few sample cutouts showing the problems mentioned. If you process some of the samples included, please specify what have you done on them (what filters, what software etc.)


  • \$\begingroup\$ Maybe you could specify whether the examples are 1:1 crops from original images, or the results of upscaling? \$\endgroup\$ Commented Feb 21, 2015 at 20:14
  • \$\begingroup\$ They are 1:1 crops from original images, stored using the maximum quality jpeg compression. \$\endgroup\$ Commented Feb 21, 2015 at 20:53
  • \$\begingroup\$ My two cents: don't bother trying to upscale them. It will be a lot more work than everything else you want to do and the results won't be great. The important things you'll want to fix: noise, white balance, distortion, and chromatic aberration. \$\endgroup\$ Commented Mar 3, 2015 at 21:59

3 Answers 3


Based on junkyardsparkle's pre-processing (I cropped the sample in order to fit 1:1 /when upscaled/ to page) I've tried to employ 10 various upscaling methods (including very exotic ones) to find out which one would cope with the weird Epson PhotoPC 600 pixel rendering best. The samples are upscaled to 200% as requested with no further post-processing.

The following results are ordered from the worst (up) to the best (down) by my own personal preferences.

10. SmartEdge 1.2 demo employing NEDI and fractal sharpening produces usually excellent results on downsized images but completely fails here producing very unnatural lined edges.

9. EANBQH - exact Area image upsizing with Natural BiQuadratic Histosplines - an upscaling algorithm with results usually a bit sharper than Lanczos a=3 fails here, too.

8. Image Analyzer - Fractal Interpolation PlugIn, Fractal9/XinLi - fractal interpolation based on Wiener filter, surprisingly almost the same results as 9.

Fractal Imager
7. Fractal Imager from Iterated Systems (1996) was a fractal image compression program (once quite famous) offering decompression to different images sizes. AFAIK the algorithm was later transformed into Genuine Fractals (see below). Still not very cool results and a visible loss of details.

BenVista PhotoZoomPro demo
6. BenVista PhotoZoom Pro demo seems to produce "ultra-smooth" edges on regular downsized images but doesn't help much here.

Yuval-Fisher QuadTree
5. Yuval-Fisher quad-tree fractal image compression sample from 1990's modified (by me) to upscale pictures by decompressing to higher resolutions. Better than 7. - details are preserved better but some artifacts are also present.

Gimp Bicubic
4. Gimp Bicubic implementation - the weird lined edges seem to be partially suppressed.

ImageMagick Gauss
3. Image Magick with Gauss filter - even smoother but a bit blurry.

Genuine Fractals 4.1
2. Genuine Fractals 4.1 (an old version) - now Perfect Resize. This is already acceptable. The images are sharp while their edges are almost smooth and without weird lines.

Gimp Lanczos3
1. Gimp Lanczos3. A very surprising winner! The edges are really smooth while the sharpness is acceptable.

As a conclusion, Gimp Lanczos3 seems to be the best option for this particular image. The results indeed don't look like from a current high resolution camera but when displayed on a larger screen it's better than relying upon an image viewer fast scaling.


While you won't get more detail or honest contrast out of the pictures than what they contain, what you can do without causing other problems is some light removal of chroma noise and other color artifacts like purple fringing, then fix bad white balance (within reason), and possibly additional subtle improvement of skin tones if needed. I did this with one of your example pictures to illustrate. It's a subjective thing, but for me this makes the biggest difference in my enjoyment of old family photos, as color tones seem to carry more emotional weight than a mere lack of contrast or detail... my two cents.

Chroma noise and purple fringing reduced, white balance adjusted.

  • \$\begingroup\$ Could you possibly specify what filters (in what software) have you used to mitigate colour related problems? \$\endgroup\$ Commented Feb 21, 2015 at 21:14
  • \$\begingroup\$ I tried to keep the answer non-software-specific, but for the example I used darktable, specifically: defringe module (desaturates chromatic abberation colors along high-contrast edges), non-local means denoise (set to operate primarily on chroma, to avoid losing detail in the luma channel), white balance (used the auto setting, although my tastes for this picture might have been for slightly warmer). One additional (and less straightforward) step was the slight shifting of skin tone hue away from green using the "color zones" module, to a degree just barely perceptible. \$\endgroup\$ Commented Feb 21, 2015 at 21:34

Modern programs can remove chromatic aberrations, increase contrast, remove noise and upscale the image. It can all be done. The problem is that these operations will decrease the resolution, introduce banding and other nasty problems.

I doubt that you would be a lot happier with the result even if the most sklled retoucher worked on your images since the prerequisites aren't there. Early digital cameras produced really bad quality images and had a far too low resolution.

  • \$\begingroup\$ I think the only operation that actually decreases the "resolution" is the upscaling. Many upscaling methods try to reintroduce details (e.g. using fractals) but it usually looks rather unnatural unless applied only slightly. Noise reduction may destroy details (textures etc.). \$\endgroup\$ Commented Feb 21, 2015 at 11:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.