Review: Optipix

[Doug Nelson]

View here

[Stephen M]

Hi Doug, thanks for the new review.

I have recently 'discovered' this software company and it's products, including two free plugs - (http://www.retouchpro.com/forums/sho...&threadid=4015).

I would like to add some comments to your review, as many users have been doing similar things with other tools or with native software features in Photoshop.


* Blending Exposures -

PC HDR Shop - High dynamic range and high bit software which can blend multiple exposures. Although intended for Photography, I have tested this with multiple flatbed scans 'exposed' at different gammas ranging from 1.0 to 3.0 with 11 0.2 gamma step scans...very slow...not for the high volume production setting, buf if you have time and like an unclipped histogram...then you get a nice high bit, 'high range' slightly flat file...which is pretty good for some basic tonal editing to polish things off.

http://www.debevec.org/HDRShop/main-pages/intro.html

http://www.debevec.org/HDRShop/tutorial/tutorial3.html

http://www.debevec.org/HDRShop/tutorial/tutorial2.html

http://www.debevec.org/HDRShop/

http://www.lightwave3d.com/tutorials/fullprecision/index.html#1

* Image Averaging -

Mac or PC free 'Panotools' plug (probably needs a review Doug, this is very deep and complex stuff, but hey - it's free). Look into the 'PTAverage' sub utility.

http://home.no.net/dmaurer/~dersch/Index.htm

http://home.no.net/dmaurer/~dersch/average/average.html

Ian Lyons tutorial on multi scanning

http://www.rgbnet.co.uk/ilyons/multiscan/multiscan1.htm

* Auto Contrast -

Photoshop Level/Curve adjustment layer set to Luminosity blend mode perhaps? There is also the auto contrast command and the new Photoshop 7 auto colour options.

http://members.ozemail.com.au/~binaryfx/PSTV_links.html#C

* Nudge -

Photoshop's filter/other/offset...nudge the object one pixel and then use the fade command to fade the effect (blur, anti-alias)? Since you can't move in smaller increments than a pixel, you have to fool the eye by averaging tonal values. Or fade the move tool etc.


* Edge Enhancement -

Many users incorporate an edge mask into thier USM sharpening tricks. A selection that hides the edges with a small smart blur perhaps...then a inversion of the selection to isolate the edges with a unsharp mask etc. Many scanners offer grain smoothing as part of the USM sharpening process.

http://members.ozemail.com.au/~binaryfx/ST_utilitymasks.html

* Safe Sharpening -

Avoiding colour noise sharpening...sounds like the good old luminosity blend or fade when sharpening - or only sharpening the L channel of LAB or B or HSB etc...or is something else happening here?

http://members.ozemail.com.au/~binaryfx/PSTV_links.html#S

It would be interesting to see how the Optipix filtering compares to the slower or clumsier hacks that are usually performed with native tools. Having a plug streamline things can be fantastic, while others may have the luxury to enjoy doing things in a more manual process...in other cases the hacks may be better than the software (Photoshop's basic toolset is very good and offers many possibilites, if only you know how to link them all together).

Stephen Marsh.

[jcr6]

Notes about Optipix

There are a number of things that we did in Optipix that I need to point out.

1. We looked into automatic image alignment and it's pretty simple if the camera (or the film on the scanner) doesn't rotate, but automatic alignment can be a bit of a trick. It gets worse if something in the field of view changes -- like coulds, a car, etc. You try to match up the stuff that moved as well as the stuff that didn't. Not simple. Some day, hopefully.

2. We thought we did a pretty good job with Blend Exposures. Look at our treatment of highlights and shadow. I'd say that being able to handle up to 6-stops changes everything about how you use your digital camera. That bracketed capture mode suddenly seems a lot more useful, doesn't it? The author of that plug-in is my father: Dr. John Russ, author of the Image Processing Handbook. He did a considerable amount of research in finding a way to do it right. If you convert both images to 16-bits first and do the blend, you can then do a useful curves or levels adjustment, because your precision has been maintained.

3. Image Averaging. We do it in a completely lossless way. I have a 32-bit per channel offscreen buffer that even works with selections that you can add a mixture of 8-bit, 24-bit, 16-bit, and 48-bit images/selections together. Since it is a Photoshop plugin, you can even sum up really dark video clips in Premiere and get a useful image out of it. I try to deliver value for the money!

4. Auto Contrast. Without loosing some bits by switching to L*a*b, doing an autolevels, and coming back (incidently something you cannot do in 16-bits per channel) how do you propose to do this in Photoshop without screwing up the color balance? The build-in Autolevels does each channel separately causing all manner of color problems.

5. Nudge. I find it very useful to have +/- 1/3 shifts programmed in on the function keys. "Fade the move tool?" Go for it. it is very useful to record a specific move distance into an action. I made this tool for myself. It seemed like a natural to include in the package.

6. Edge Enhancement. Y'all missed the boat on this. We're not doing an Unsharp Mask (which is technically a Laplacian of the Gaussian). We're doing a Difference of Gaussians which lets you do a bandpass, not simply amplify the frequencies right below the HF cutoff. In addition, by letting you pick either the light or the dark edges (a way-cool thing to do with portraits) you can get the advantages of sharpening detail without making nasty halos around them.
http://reindeergraphics.com/examples/testedgebefore.jpg
This is the original image.

http://reindeergraphics.com/examples/testedgeboth.jpg
both light and dark edges - similar to unsharp mask but less noise emphasis

http://reindeergraphics.com/examples/testedgedark.jpg
Dark edges only

It really does make a difference. We're not just blowing smoke.

7. Safe Sharpen is here because it works in 16-bits. Yes, it is similar to the L*a*b processing that you describe (with some attention to chroma noise), but you can do it without loosing any precision in your images?

The whole point of Optipix is to preserve and optimize the content of your pictures. I want to save every last little bit so you have more options down the road in your processing. If you're doing even a few steps of processing to your image, 8-bits per channel isn't going to cut it.

(end rant)

-Chris Russ

[Doug Nelson]

In my own defense, on page 13 of the user manual it says "Edge Enhancer is a sophisticated Unsharp Masking tool".

[jcr6]

Fair enough. George's words. (It's hard to explain the difference between DoG and USM.)

Perhaps I should delete my rant from above.

Also, the manual didn't include the changes for 1.0.2.

So, what kind of plug-in should we write next? The scary but necessary image alignment plug-in?

-Chris

[Stephen M]

Re: Notes about Optipix

Hi Chris, thanks for taking the time to further inform us on the abilities of your product. It can be a very rewarding experience for both end users and software developers when they have frank discussions concerning each of their respective issues and needs. Keep in mind that I am just a end user with no programming knowledge (apart from Basic on the Atari when I was a kid <g>) and the closest I have come is using custom convolution filters, let alone filter facotory or more serious plug creation.

Firstly - I think your company should seriously consider a demo version of the plugs of some sort (time limited, watermarked, pixel limited, preview only etc). Many users need to get a feel for things in person to see if the plug is for them - reviews and vendor information are not enough. I guess that I am not in the direct target market for your product - but prepress often has similar concerns as photographers. Another factor is that the more advanced a user is, the less they feel the need for third party add on's - when they know how to exploit Photoshop's numerous abilities. That being said, when a utility is found that meets or exceeds expectations, then the user is very happy with the purchase and is often a good word of mouth advocate.

There is a lot to go over, so I will respond in small chunks to each of your replies to my original musings.

There are a number of things that we did in Optipix that I need to point out.

2. We thought we did a pretty good job with Blend Exposures. If you convert both images to 16-bits first and do the blend, you can then do a useful curves or levels adjustment, because your precision has been maintained.

Yes, this is a nice feature - although not unique to your product it is nice that you offer the ability to merge multiple high bit files, which many photographers like using.

I am not sure what your point is on moving an original 8 bpc file to 16 bpc before averaging...as no new valid data is generated - unless the averaging process yields a better result when performing in/on high bit data.

Are you saying that 8 > 16 bpc > Blend Exposure yields a better result than 8 bpc > Blend Exposures? How about a true high bit file that does contain valid extra data? If the false high bit file is better due to the calculations done in high bit, then having true extra data should be a significant improvement again.

90% of my work is presented as CMYK 8 bpc data direct off a high end Crossfield PMT drum scanner - and all of my edits are performed in/on 8 bpc data...All of my edits seem useful to me! <g> I think the supierior 8 bpc data off the drum beats any 16 bpc data off a CCD flatbed, when it comes down to useful edits...but I could be wrong.

Even if I moved the data up from 8 to a false 16 bpc - can your product work with CMYK files? If you have to convert from CMYK to RGB to use the plugs - then I would think that any concern over bit depth would be misplaced.

Is the better result visible to the human eye as an improvement to the appearance of the image (that is worth the extra time/effort), or is the benefit only indicated in the histogram? And if the effect can be seen at 1:1 size on the monitor - will it be seen on final common print output to inkjet or press?

Some users care about the data, others may only be concerned with the final visual results and may not care about extra precision behind the scenes - if it can't be directly billed to the client (high bits do take extra time and graphics is not a hobby for all users). Most clients only care about how their image reproduces, not how many levels or useful bits it has (although some who purchase data do care about these factors).

[4. Auto Contrast. Without loosing some bits by switching to L*a*b, doing an autolevels, and coming back (incidently something you cannot do in 16-bits per channel) how do you propose to do this in Photoshop without screwing up the color balance? The build-in Autolevels does each channel separately causing all manner of color problems.

For 8 bpc images, an adjustment layer set to luminosity blend or a fade command to luminosity mode would be the preferred edit if not working the separate channels in normal mode to gain colour and contrast. Although not the same as L of LAB, this luminosity average blend and it's opposite (color [hue+sat]) can often do similar things to LAB and HSB modes without the very minor visual penalties of some mode changes (which may or may not be common).

Photoshop offers a basic auto contrast option - which affects the levels/curves differently to auto levels/curves - although it is not doing a luminosity blend (you can add a luminosity blend/fade to the auto contrast command). Are you familiar with luminosity/color blends and the contrast and the new level/curve options of Photoshop 7?

As for high bit images - I agree that Photoshop's current inability to layer 16 bpc data makes the use of luminosity blends a dead issue.

[5. Nudge. I find it very useful to have +/- 1/3 shifts programmed in on the function keys. "Fade the move tool?" Go for it. it is very useful to record a specific move distance into an action. I made this tool for myself. It seemed like a natural to include in the package.

Not sure here Chris - unless you can defy the basics of raster graphics, the nudge plug will be using some sort of averaging technique to simulate the sub pixel location of the new pixel location. I once had some custom filter kernels for this, but I have lost them (David Biedny had them on offer on the dead PhotoshopCentral site years ago).

Unless I am mistaken, using the offset filter or the move tool and nudge keys to *fade* a single pixel movement would provide a similar result - although the underlying maths may be different.

I am interested in why you recommend 1/3 increments - I would have thought that even increments would have been better, such as 1/4 or 1/2 pixel steps?

[6. Edge Enhancement. Y'all missed the boat on this. We're not doing an Unsharp Mask (which is technically a Laplacian of the Gaussian). We're doing a Difference of Gaussians which lets you do a bandpass, not simply amplify the frequencies right below the HF cutoff. In addition, by letting you pick either the light or the dark edges (a way-cool thing to do with portraits) you can get the advantages of sharpening detail without making nasty halos around them.

It really does make a difference. We're not just blowing smoke.

OK. This is why I wanted to play with KPT Convolver, but there is no demo. For serious exploration, you need quick interactive feedback via a GUI control, which the custom filter does not offer. Thank you for offering some basic DOG txt files in your free custom filter, as this will let me explore this concept to see how it differs from USM. Your product adds my requested GUI component to the DOG operation, which is great.

As for the split light/dark halo concept - this is nice to see. Many higher end scanners offer many sharpening options such as both smoothing and sharpening - as mentined above. They also offer the ability to control the % of dark or light halo intensity. It seems that your product offers only on/off for dark/light halos? It would be a major benefit to have sliders to control the application of dark/light haloing. Often you want the dark halo at 100% but need the lighter halo reduced in intensity, but not absent.

All of this can be done in Photoshop with layers or fades, but not on high bit files, although they can still be sharpened.

[7. Safe Sharpen is here because it works in 16-bits. Yes, it is similar to the L*a*b processing that you describe (with some attention to chroma noise), but you can do it without loosing any precision in your images?.

For 8 bpc data this can all be done without loosing any precision via a mode transform using blending modes/fades (math may be different, but the visual results are usually close). If you mean can it be done on high bit data, then again - no, not with Photoshops current limitations on layering/blending...although with history states there are some possibilities...

[The whole point of Optipix is to preserve and optimize the content of your pictures. I want to save every last little bit so you have more options down the road in your processing. If you're doing even a few steps of processing to your image, 8-bits per channel isn't going to cut it.

(end rant)

-Chris Russ

OK - I agree in theory and concept, but the real world results are usually lacking and contradict the theory. I will try to use high bits in acquire (scanner, camera)...but for edits in Photoshop things are not so clear cut.

My situation demands a practical mindset. Time is very limited and there are a large amount of images to process. Most useage of the images is one off for offset press on good matte or glossy stock. If the final printed results do not justify the extra time and hassles that high bit data or workflows add - then it is hard to implement thier use.

I would not have a job if I missed magazine press deadlines because my edits took twice as long becuase I was working with high bit data instead of regular bit data, when the images often do not show the great benefit at the end of the day when the final prints are on the stands. When processing power and editing software allow full flexible use of high bit data in real time with no hits to productivity, then I have no problems with this workflow concept. Until then, high bit edits are more of an interesting theory than standard practice.

I would have to say that my edits *do* cut it, even though they are on 8 bpc data - but this is a subjective rating after all. <g>

Thank you for the exchange Chris, I have enjoyed the exchange and gained some deeper knowledge on image processing, I also hope that you can gain something from my 'in the trenches' perspective too.

P.S. I do not consider this a rant at all, these are deep and serious topics that are not easy to comment on without boring the author, let alone the audience.

Sincerely and respectfully,

Stephen Marsh.

[Stephen M]

Deblur, Deconvolution etc

Chris - to quickly answer your question on what software you should develop next...

I notice that you have a De-convolution filter in your other products, FP + IPT?

If you could market a Photoshop plug for Mac/PC to bring blurry or out of foucs images into focus, that would be nice. <g> Undoing motion blur over one or more directions/dimensions/amounts would be nice too.

http://www.focusmagic.com/index.htm

http://rainier.medinfo.rochester.edu/~wodell/courses/BMEgradLab_fall01/ImgProcOutline1.html

http://www.ph.unimelb.edu.au/~jpc/pages/unblur/samples.htm

http://www.indiana.edu/~elegans/Deconv/DeconAppNotes.html

Also, I understand that Fourier filtering may offer some advantages? Is there something you could do there (I think your other products have this too?).

Regards,

Stephen Marsh.

[jcr6]

What we were thinking

About a demo version... With FoveaPro 2.0 and the Image Processing Tool Kit 4.0 we made a time-limited demo, largely because we had an established product and we understood the market that it was in enough to justify the time to make one.

In fact, that's the reason we didn't with Optipix 1.0. We had no idea if anyone would care if we had the product or not so we decided to save the engineering time and not make a demo. If people were interested, we could come along and do a demo on the next version. And we probably will. It is also easy to say "Make a demo" and a good deal harder to write one.

First, you have to have a method of security that makes the software somewhat less useful, while at the same time gets the point across what it can do. Personally, I hate watermarking images because then you can't do the rest of the steps in your workflow to see how the software will work. That leaves timing-out which is just as difficult as creating security. How do you prevent people from faking out the time limit? It takes engineering time, and if it is a really good plug-in, people will take the time to break it and spread the secret around. Can you say "piracy?" Gee, I knew you could.

We do have some free plug-ins and you can learn a lot about us from them, from our other products, from our reviews, and from our customers.

I've been working on Image Processing and Image Analysis since 1981 on the Apple ][.

<Aside: Steve, you've made it pretty clear that you're not in the target market. That's cool. I think we should start another thread about plug-ins that you might be interested in.>

I wasn't trying to poke holes in your ability to edit an image in 8-bits. I was trying to say that with successive processing steps the number of real bits in an image tend to dwindle due to truncation, round-off, and mode-changes. If there wasn't that much to begin with (say from VIDEO), life isn't good.

I think we should talk about the merits of CMYK vs RGB (vs L*a*b vs CIE vs HIS) in another topic. I would like to talk about that, along with human perception and some other things. Among other faults, I look at a picture as something that a scientific instrument took, where the instrument is capturing real detail, even if I cannot see it, and that point-of-view is much different than the photographer or artist.

FWIW, most of our processing takes place in L*a*b space (with floating point numbers to preserve the color values). Sometime, look in the blue channel after an image has been JPEG'd, or better yet, do a transform into HSL space and see how BAD the image is.

1/3 nudge vs. 1/2... Make an image of vertical lines - even pixels black, odd pixels white. If you shift it by 1/2 a pixel, you get gray. If you shift it by 1/3 a pixel you still have lines. Shift it by 1/3 again, you still have lines, shift it by 1/3 again, you still have lines. This is caused by working right at the Nyquist frequency (the maximum resolution in the image) where there is no place to store the information. Thus, you can shift by an odd fraction of a pixel and preserve most of the information, but not by 1/2.

Edge Enhancer light/dark halo slider controls: I must say that I didn't even think of it, but the dialog is too busy already. Something to think about. Thank you.

As you say, Optipix was more oriented toward Photographers than Graphics Artists, but we thought (HOPED) that enough of the tools would be worth it to y'all that you'd be interested in it, too.

Let's start another topic called "Plug-ins for Graphics Arts" and bring our thinking caps. I'm agressive in pursuing beta testers.

-Chris Russ

[Stephen M]

Hi again Chris, thanks for the reply.

On the Optipix demo - I think that is something you should look into, there are many native techniques and competing plugs on the market and it can be hard for users to know if they are purchasing something worthy or just something that sounds like it might be worthy. No biggie - but I think it may be a positive thing to introduce, if you can solve your technical concerns with the demo that is.

Agreed on the loss of levels in 8 bpc edits, it is just that the final use of the data that I am concerned with rarely shows any benefit for use of higher bits for the extra processing they required. With computers becomming faster, this may become less of a concern for production orientated people as time moves on.

Yes - there are both 'inputcentric' and 'outputcentric' extreme viewpoints, and many shades inbetween. We look at things from the opposite viewpoints - which is not a shock, as our respective useage and expectations of data and image processing are often radically different.

Thanks for the explanation of the 1/3 vs 1/2 or 1/4 pixel nudge offset...food for thought indeed. My mistake on the move tool/cursor key/fade combo...no such beast, but I did quickly play with fading the offset filter - but I can't compare my results to anything...perhaps your Optipix page needs a visual example in GIF format of the difference between the nudge and the Photoshop hack.

On the Optipix edge enhancer - users can work around the lack of variable halo intensity control this using layers/blends/opacity or fades to lighten/darken mode, or even blend if layer sliders. Users in high bit mode can also do the same in a more convoluted process using history brush and states if I am not mistaken (but I have not tested this theory).

Thank you for the discussion Chris, it has been fun and I have learned a thing or two.

Sincerely,

Stephen Marsh.

[jcr6]

Food for thought.

Actually, Steven, you made me think about different points of view.

We were really trying to point Optipix toward people who had a lot of images and needed workflow solutions, while at the same time trying to provide something fairly unique.

You're a different case (I wouldn't say a special case 'cause I'm not evil ), and asked some hard questions. I reacted badly (as many programmers tend to do when their "children" are under scrutiny). Sorry about that.

So I would like to talk about the tools YOU think are missing. What do you spend a lot of time spinning you wheels doing that could make your time more valuable?

Yes, there will be an Optipix Demo for 2.0. No question.

Actually, you can do an opacity fade with a shifted layer and achieve a fairly decent nudge effect, but you don't have a lot of flexibility doing it. (You made me go look.) The results are similar, except for boundary conditions. You will get better results by working in superresolution (zoomed up by 2x or more) because there is no way to represent alternating lines shifted by 1/2 pixel.

And we've beaten the other topics to death. You have demonstrated that you understand what is going on INSIDE of Photoshop to a very high degree. To that end, I think you (and others reading this post) would probably make excellent beta testers and a souce of constructive criticism.

So what's next?

-Chris

[Stephen M]

1/2 &amp; 1/3 Pixel Nudge Custom Convolution

An update - I lamented that I had lost the 'sub pixel motion' convolution kernels that were from a defunct website...but with a very extended search and quite a bit of luck - I have found what seems to be an online 'archive' of the site, even though it is no longer running (or appears to be no longer running).

Anyway - the links still work (for now) and you can download the custom convolution kernels here:

http://macanudo.jpq.com:8088/idig/photoshopcentral/tips.html

http://macanudo.jpq.com:8088/idig/photoshopcentral/tips/mo_kern.zip

http://macanudo.jpq.com:8088/idig/photoshopcentral/tips/mo_kern.sit

With the recent EXCELLENT posts from Chris Russ on custom convolution (see link below) and his great free custom 'custom' convolution filter I will shortly experiment with these kernels with more understanding than in the past (thanks Chris), I just wanted to share the links first!

http://www.retouchpro.com/forums/showthread.php?s=&threadid=4015

Looks like I'm turning into a convolution geek - when I am a diehard GUI person. <g>

Stephen Marsh.

[Stephen M]

Quick Test Results of Sub Pixel Motion Convolution:

Rasterized EPS font, 3pt black Times text and also inverted (very small fine serifs) -

Did a offset filter of 1 pix and a 50% opacity fade and compared it to the custom filter convolution kernel for 1/2 pixel...

The custom convolution method is the winner - sharper serifs and fine detail (which the fade command with the offset filter blurs slightly more), although printing may not show the difference...it is probably a big deal though when stitching or aligning images, which is probably where Chris is coming from with his third party nudge filter in the Optipix plug-in package.

Stephen Marsh.