Sharpness/Resolution/Resizing/Confusion

[keliot]

Hello.
I am getting conflicting and confusing information regarding resizing software being able to increase resolution of the image. If the image is resized up and "new detail" is extracted but then downsampled then the resolution will increase and will this new detail now make an image even sharper?

Also, the image resizing software claims that they use the deconvolution methods which is the same as what the sharpening sofware advertizes as using.

I am looking at Piccure and Sharpen.

Looking to extract maximal detail and resolution/sharpness/focus beyond the traditional sharpness tools such as Topaz or Nic or FocusMagic, etc.

Thank you.

[skoobey]

Re: Sharpness/Resolution/Resizing/Confusion

No resized image will ever be more detailed than the original. That's a fact.

[Tony W]

Re: Sharpness/Resolution/Resizing/Confusion

It is hardly surprising that you are finding the claims confusing -IMHO there is an awful lot of BS spread about this subject.

Claims made by the purveyors of software are rarely met in practice, at least to the amount one would hope for after drinking their particular brand of Koolaid.

• Initial best bit of advice is lower your expectations quite a bit from the claimed superiority of one vs another.
• Second try out the demo programs on some of your images and compare to the best you can get out of your existing software.
• If you cannot get good sharpening out of LR or PS then there is something wrong with your technique.

It is rare that you can polish a turd to a high shine so if your images are not first class to begin with then sharpening methods are really not going to be of much use, at least to get a first class image. Deconvolution algorithms attempt to reverse blurring by reconstructing the blur rather than just applying light and dark edges as with conventional sharpening - but there is only so far that they can go to this end

I have tried many sharpening programs (including the ones you mention) and TBH found nothing outstanding in comparison to sharpening using the tools in LR, ACR and Photoshop.

It often gets overlooked that in LR, ACR and PS deconvolution sharpening is available. LR and ACR apply a mix of USM and Decon. depending on the amount setting for the Detail slider.

Up to about 50% there is a combination of USM/Decon. After 50% it is pretty much all deconvolution. Photoshop Smart Sharpen when set to Lens Blur uses decon.

Focus magic, Topaz Infocus also use deconv.

Raw Therapee (Free) also uses deconvolution and several algorithms and iterations to choose from.

[Tony W]

Re: Sharpness/Resolution/Resizing/Confusion

No resized image will ever be more detailed than the original. That's a fact.

Not quite true if the original image is mildly blurred and the PSF required for exact deconvolution can be found the image losses can be mitigated and restored to a degree.

These algorithms were used to bring back some of the losses to the originals incurred with the Hubble telescope incorrectly ground mirror which caused blurry images

https://www.nasa.gov/sites/default/files/thumbnails/image/gpn-2002-000064-crop_hubble_images_of_m100_before_and_after_repair.png

http://hubblesite.org/gallery/album/entire/pr1993010a/large_web/

[Shoku]

Re: Sharpness/Resolution/Resizing/Confusion

While I agree with Tony W, I have found Ben Vista's Photo Zoom Pro to be the best for enlarging good quality photos more than 150%.

From the website:
"BenVista's unique S-Spline Max technique is specialized in creating sharp, crisp clear, and true to nature image enlargements. Both strong edges and subtle details are being preserved efficiently. In addition, S-Spline Max is able to reconstruct the shapes of portrayed subjects in a very refined way, allowing them to keep looking natural in enlarged form."

The program offers multiple ways to enlarge images and beats standard Photoshop algorithms.

http://www.benvista.com/photozoompro/examples

[skoobey]

Re: Sharpness/Resolution/Resizing/Confusion

Not quite true if the original image is mildly blurred and the PSF required for exact deconvolution can be found the image losses can be mitigated and restored to a degree.

These algorithms were used to bring back some of the losses to the originals incurred with the Hubble telescope incorrectly ground mirror which caused blurry images

https://www.nasa.gov/sites/default/files/thumbnails/image/gpn-2002-000064-crop_hubble_images_of_m100_before_and_after_repair.png

http://hubblesite.org/gallery/album/...10a/large_web/

That's why i wrote the "more" part. Yes, it can look unnoticeable, can it gain more detail? No. What's there is there.

[klev]

Re: Sharpness/Resolution/Resizing/Confusion

I have no idea why these developers use terms like convolution and deconvolution when most people have never been exposed to those particular concepts.

Convolution refers to something called an integral over the product of two functions. You know that you can multiply numbers. You can do something similar with functions, even though it's not always as you might imagine it. Deconvolution is just an inverse operation to that.

These software packages work with numerical data, which results in an error value that is compounded with each operation on the data. They can't perfectly invert an arbitrary operation due to the way floating point arithmetic works, so you can't even something as good by upsampling and downsampling again. This is particularly true when the high frequency information tends to go first, and you have no way to recover that.

Shoku's referring to something a bit different. There have been many attempts to deal with things like object removal and inpainting. It was an area of active research, and it's mentioned frequently in one of the more popular overviews of outstanding problems in image processing. I'm linking that solely for the descriptions. Don't pay attention to the math symbols. Even if you recognize them, many of them mean something very specific in this context.

I forgot to write a couple things. Inpainting is accomplished in a few ways, but they mostly look at surrounding values and discrete approximations of the rate of change approaching each point. Beyond that they're often performed in an iterative fashion. It's not like you get something better or even as good as you would normally. It's just that some algorithms are sufficient to produce something acceptable.

[Tony W]

Re: Sharpness/Resolution/Resizing/Confusion

That's why i wrote the "more" part. Yes, it can look unnoticeable, can it gain more detail? No. What's there is there.

No, you are at best missing the point at worst just plain wrong. In short, yes you can gain more detail due to reversing the blur created in the original by such things as missed focus and diffraction, but there are of course limits - no CSI stuff as yet.

Conventional sharpening algorithms such as Unsharp mask apply a light and dark halo around parts of the image giving the impression of added sharpness. This being accomplished by a combining a blurred negative image with the original. The technique predates digital by a wide margin 1930's.

Deconvolution algorithms operate very differently and are much more complicated and require intensive processing leading generally to longer times to complete.

The goal of deconvolution is to recreate the signal as it would have existed prior to the blurring (convolution) caused in the image by whatever means. This is of course very different to just applying edge halos as USM and is probably more fitting to apply these algorithms (decon) to a raw image in the first stage of a sharpening regime (forget JPEG they have already been sharpened and you will do more harm than good)

I have no idea why these developers use terms like convolution and deconvolution when most people have never been exposed to those particular concepts.....

WTF ?
This part very simple to answer, the developers feel the need to differentiate between conventional USM sharpening and the fact that they are using algorithms that are doing something that USM cannot and in a very different way.

You seem to be overcomplicating something that is already complicated. IMO it may be more digestible to most if we just say that from a photographic POV that an image that is unsharp is said to be convolved and that to attempt to put this right we need to deconvolve the image. Thereby improving the resolution of the image data by mathematical algorithm designed to separate information from artefact.

There are many different decon. algorithms including Richardson Lucy, Weiner etc. each claiming particular benefit. Raw Therapee uses Richardson Lucy and offers filter dampening and you can choose the number of iterations! I suspect that the so called unique S Spline algorithm is one including decon. at least in part.

An enlargement of 200 - 300% is fairly trivial task at least within PS or LR and while the latter is more limited due to the lack of adjustment layers and blending the results are generally far from not being acceptable and even better when one considers a proper viewing distance for a print.

I have yet to see a truly compelling demonstration of enlargement software that really adds much to the IQ table from that which can be achieved in Adobe products. I would love to such but would insist in being provided with the original image data and also the processing steps taken so that my own comparisons could be made - otherwise we are looking at subjective and biased views rather than looking objectively

I have seen plenty of advertising blurb showing a companies (or individuals) comparison between 'ours and Adobe', but interestingly enough they never seem to allow you access to their original image to enable you to try yourself - I wonder why?

[Shoku]

Re: Sharpness/Resolution/Resizing/Confusion

I have seen plenty of advertising blurb showing a companies (or individuals) comparison between 'ours and Adobe', but interestingly enough they never seem to allow you access to their original image to enable you to try yourself - I wonder why?

Looks like Photo Zoom Pro gives you the original images, which you can adjust by choosing the process you wish to test from the list on the left, as well as add or remove other options.