Help optimize sharpness in small image

drcarl

I need to learn how to maximize sharpness on images I reduce in size and dpi.

I am shooting images of products for display on a website. After I shoot, I isolate the product with the lasso tool photoshop, use some unsharp mask, and then shrink them with "bicubic sharper" set...then I save them as a .png.

I don't know if I am doing something wrong, if I am missing something, or if there is something else I can do to keep the images from going fuzzy on me. I might even be expecting too much...and feel a bit like I am re-inventing the wheel since there are SO many variables.

I'll list my steps and some samples with the great hope that someone can help me improve my techniques.

Thanks in advance,

drcarl


DETAILS

Product bottle in tent lit with fluorescent 'curley' floods
Camera Canon 10D, 24-105 L series zoom, stabilizer off, tripod
1/4 of a second, f22, zoomed at about 45mm, ISO 200, sRGB color space
original image size 382 KB

http://drcarl.smugmug.com/photos/340615783_UBijG-M.jpg<o:p></o:p>

selected bottle with lasso tool<o:p></o:p>
inverted selection<o:p></o:p>
removed background with background eraser<o:p></o:p>
cropped close to image<o:p></o:p>
<o:p></o:p>
Used automated batch file to….<o:p></o:p>
open Image Size<o:p></o:p>
unchecked resample image<o:p></o:p>
changed 180 pix/inch to 72<o:p></o:p>
clicked OK<o:p></o:p>
<o:p></o:p>
opened Image Size<o:p></o:p>
checked resample image<o:p></o:p>
selected bicubic sharper<o:p></o:p>
changed width from 1536 pixels wide to 200<o:p></o:p>
clicked OK<o:p></o:p>
<o:p></o:p>
applied unsharp mask 32%, 1.6 pixels, threshold: 2<o:p></o:p>
saved as .png

Opened all images and adjusted levels for art and continuity
(black, white and the middle slider)
re-saved<o:p></o:p>

http://drcarl.smugmug.com/photos/340615858_FwxVH-X3.png

Now, since the image needs to be about 50 pixels wide, with bicubic sharper selected, I reduced it again to 50 wide.<o:p></o:p>
<o:p></o:p>
But, it’s now too fuzzy!
http://drcarl.smugmug.com/photos/340616173_LZXva-X3.png

So I tried selective sharpening before shrinking it to 50, I tried it after. I tried the unsharp mask again before and after. I even over-sharpened it FAR more than looked right to see if, once at the teeny size, it might look better. It does, a little.

This is where I gave up and seek your expertise. Help!<o:p></o:p>
<o:p></o:p>

I know there are many ways to skin this cat, and that there are methods that will yield the best results possible. I am shooting the smaller file size initially and not utilizing all 6 MP because I figure it’s going to be teeny anyway. I hope that is right.<o:p></o:p>
<o:p></o:p>

Can anyone correct my ways? Or, am I asking for too much performance from the image at this small size and how its pixels mingle with those on the monitor?<o:p></o:p>
<o:p></o:p>

Here is a link to all my variations as well as a shot of the header my client is making. She’s using a .bmp, but since that file type is unsupported here, I saved it and posted here and in the gallery as a .jpg and a /png.<o:p></o:p>

http://drcarl.smugmug.com/gallery/5559486_cEzLc/1/340616173_LZXva

Right now it’s close, but no cigar.<o:p></o:p>
<o:p></o:p>
Thank you for your help.<o:p></o:p>
<o:p></o:p>
drcarl<o:p></o:p>

PS - I wish I had ALL the perfect steps...step-by step

BinaryFx

drcarl, I will have to come back to your question on retaining the appearance of detail when reducing image size to very small pixel dimensions.

Do you notice the light coloured edge halo around the product?

Once you have permanently removed background pixels to transparency, you can use the layer matte command to clean up the colouring of the edge pixels (you are working on a duped copy of the original, right?).

Below is an example of using the defringe command at 1 pixel (layer/matting). Sometimes remove white matte (or black) will be OK, some other times defringe or combining multiple methods will be required.

Anyway, I personally would use a different method to isolate the product and create a transparent background where one would have a better edge that may not reqire the matte command, but that is another topic!

The right hand side of the shot below has had the defringe command applied, while the left hand side is the original.


Hope this helps,

Stephen Marsh
http://members.ozemail.com.au/~binaryfx
http://prepression.blogspot.com

drcarl

Great!

BinaryFx wrote:

drcarl, I will have to come back to your question on retaining the appearance of detail when reducing image size to very small pixel dimensions.

Awesome...and thanks for addressing the the other issues as well

BinaryFx wrote:

Do you notice the light coloured edge halo around the product?

Yes. But I thought that this was from placing the dark bottle inside a white tent with a white background. Shooting the almost black bottle on the black background made it almost impossible to isolate, and the other color tents (red/blue) cast their own tint onto the bottle and label. I tried using the magic wand in contiguous mode but it would select into unwanted areas (like bottle text or edges almost identical to the background) so I felt left with manually selecting the bottle with the lasso. As I look more closely I see exactly what you are referring to; it's like a halo. (I also forgot to mention that I used the blur tool a tad to try to correct for this.)

BinaryFx wrote:

Once you have permanently removed background pixels to transparency, you can use the layer matte command to clean up the colouring of the edge pixels (you are working on a duped copy of the original, right?).

Wow. This is very nice. I am self-taught in Photoshop and never cease to be amazed. Defringe appears to do exactly what it sounds like it'd do. Thank you!

Yes, I am working on dupes and am even willing to re-shoot, and re-do any and all steps if that would make a more perfect image for my most wonderful, artistic, (and patient) client (who is a bit of a perfectionist like me).

Which reminds me of another question. Should I be shooting in RAW or with large files even though the final result will be these teeny tiny images?

BinaryFx wrote:

Below is an example of using the defringe command at 1 pixel (layer/matting). Sometimes remove white matte (or black) will be OK, some other times defringe or combining multiple methods will be required.

If the background is transparent, I wonder what matte, either black or white, is removed?

BinaryFx wrote:

Anyway, I personally would use a different method to isolate the product and create a transparent background where one would have a better edge that may not reqire the matte command, but that is another topic!

Oh, do tell! Ultimately, that IS the topic. I was headed for the magic wand, but just could not get proper control over it. I know there must be a right or better, more automated way. I'll be shooting some 500 products and seek the most effective and most automated workflow....that will, of course, produce the most beautiful little teeny images possible!

ANY guidance, direction, links, and ideas are heartily invited and welcomed!

Thank you SO much for your input and the example.

I wish I knew how to do it right. Bit by bit, I'll learn it.

Thank you more,

drcarl

PS - I wonder when and if I should open up and run Fireworks on images such as these???

BinaryFx

drcarl, the fringe is from the method used to isolate the product when creating the transparent background (the selection required more refinement, shrinking it into the product instead of slightly outside on the original white background). Cutting into the dark bottle a bit would not have the white halo/fringe, as would a better method of selecting/creating transparency.

If one has a true transparent layer (not a layer mask transparency) - then one can access the defringe command under the layer menu. If the original background was white, use white matte (use black matte if the original background was black). For other colours or when these two commands do not do what you require, then use defringe at 1 to 2 pixels (wider settings indicate that you may need to do a better job of isolating the image).

As for better methods to isolate, as I have been using the vector pen tool for many years - that is a fast and natural choice for me on a smooth object. From there I would make a raster layer mask with a fine soft edge from the pen path selection. The other mainstay is to find a channel with good contrast between the subject and background and to dupe and edit that for making a good layer mask. Photoshop also offers an "extract" command and a wizard/assistant to help produce layer transparency.

As for resizing...

With a 50 pixel wide reduced size image, you are going to have problems reading the text! I would think that any improvements will be minor, although I am happy to see a method that would retain legibility at this small size and be a *big* improvement over your current resize methods (although one may be able to do slightly better or worse than your example). Some may retouch in new text rendered at reduced size, if that was more legible!

I have not updated my links for a while, but if you scroll down to "masking and extraction" you will find a few, I am sure that others can chime in with their preferred tutorial or plug-in for removing backgrounds:

http://members.ozemail.com.au/~binaryfx/links.html#M (perhaps start with the graphics.about.com link)

If you are shooting JPEG at a smaller than sensor size, then you may be better to shoot at native sensor size and resample that image down. I see no benefit for raw data for a low resolution website banner where the final image is 50 pixels wide. That being said, unless you like to reshoot, there is a chance that the product will be required for print or another use at a higher resoultion, so you may be better off working from a raw file as a master. If this banner is for print and not web, then you may be working in higher resolutions anyway.


Stephen Marsh
http://members.ozemail.com.au/~binaryfx/
http://prepression.blogspot.com/

drcarl

wow

BinaryFx wrote:

...Cutting into the dark bottle a bit...

I thought I was doing that...I guess not into the product's pixels enough

BinaryFx wrote:

...a true transparent layer (not a layer mask transparency)

hmmm...now I have to figure out how to make one of these. I tried making an adjustment layer and erasing the background, but that not right. (the whit background just sat there and laughed at me - lol)

BinaryFx wrote:

the vector pen tool

I guess I need to learn THIS cool tool.

BinaryFx wrote:

...make a raster layer mask with a fine soft edge from the pen path selection, channels with good contrast between the subject and background, working with layer masks, the "extract" command and wizard/assistant...

woo! much to learn...thank you for pointing me in the right direction.

BinaryFx wrote:

...a method that would retain legibility at this small size and be a *big* improvement over your current resize methods...

IS there a better way?

BTW - thank you for the links - wow again.

BinaryFx wrote:

...shooting JPEG at a smaller than sensor size...

Just did the test...shot one at the largest size, and another at the smallest. Once reduced that much (I took them to 75 pix wide) they look identical.

And this is ~only~ for the web, not print.

Thank you for your help.

Besides being swamped with other things, and wanting to read the tutorials on isolating the product from the background, I still want to confirm that I am following the best procedure for maintaining as much sharpness as I can when reducing to 50 pix.

Thank you again,

drcarl

BinaryFx

drcarl wrote:

I thought I was doing that...I guess not into the product's pixels enough

I may zoom in to say 200-400% to make sure. Refining a mask is a basic yet critical part of the process. In CS3 Photoshop adds a new refine selection command which consolidates different separate manual methods that were employed in previous versions.

hmmm...now I have to figure out how to make one of these. I tried making an adjustment layer and erasing the background, but that not right. (the whit background just sat there and laughed at me - lol)

Let me expand a bit on this point, to clear up any possible confusion.

Forgetting adjustment layers which have a layer mask applied by default, we are talking of true pixel layers here and creating a non destructive layer mask to create transparency without permanently removing the unwanted pixels. The trade off here is that one can't use the defringe command which requires "true" (permanent) transparency, as opposed to transparency created via an adjustable layer mask.

Anyway Google around or look in any book on Photoshop, since layers were introduced in version 3 (not CS3!), I can't think of one topic that receives more coverage than layer masks. Also search for alpha channels which are related, as a layer mask is just a special alpha channel.

I guess I need to learn THIS cool tool.

The pen tool can be hard to learn, although it is well worth it (even more so if you use Illustrator).

Just did the test...shot one at the largest size, and another at the smallest. Once reduced that much (I took them to 75 pix wide) they look identical.

OK, it was worth a shot!

When shooting at a reduced size in camera, one is relying on the resizing provided by the camera software to reduce the larger true capture data to the chosen "requested capture" size. Photoshop or other software may do a better job, which is what I was interested in seeing.

Besides being swamped with other things, and wanting to read the tutorials on isolating the product from the background, I still want to confirm that I am following the best procedure for maintaining as much sharpness as I can when reducing to 50 pix.l

I think you are doing fine drcarl, perhaps there are better Photoshop techniques or preferred results available with plug-ins or other software, although you are getting resonable results considering the final size. I tried a few tricks and did not get a significant improvement over your result. I too would be interested if somebody that works with such small resolutions has any tips or better results (being in prepress, most of my processing is for high resolution, not low).

You may wish to experiment with slightly different USM techniques:
http://members.ozemail.com.au/~binaryfx/howto_sharpen.html

Or different sharpening than USM (MS Win):
http://members.ozemail.com.au/~binaryfx/sharpcontrol.html

Later versions of Photoshop offer the "smart sharpen" filter, which provides basic deconvolution (focus restoration). There are other third party deconvolution focus restoration/blur removal options that may also provide better results (focus magic, focus fixer etc).

With the new bicubic sharper resizing, most recommend a single reduction from high to low resolution. As you are finding, incremental resizing steps with sharpening before each resample may provide better results for this type of image content at this size.

Another option before/during resizing could include:
http://www.topazlabs.com/topazlabs/03products/topaz_vivacity/gallery/40create_interesting_effects/index.html
http://www.topazlabs.com/topazlabs/03products/topaz_vivacity/


Regards,

Stephen Marsh
http://members.ozemail.com.au/~binaryfx/
http://prepression.blogspot.com/

drcarl

delay
UPDATE: I Broke my ankle....this certainly caused me a delay in EVERYTHING I am up to....

Wanna see some kind of gross shots? Check out this small gallery

"How to have a broken ankle"....lol....it's still in rough draft mode.

http://drcarl.smugmug.com/gallery/5616002_5AcNa/1/344872011_8oBja

Back in a week, or two, or three....


drcarl

drcarl

Update

drcarl wrote:

Back in a week, or two, or three....

UPDATE: Well, a little more than a week. I only recently (like today) have mostly abandoned my last crutch, still can't walk freely.

The client has not touched my mini studio set-up since I fell. Wow. OK, I should say The Friend (and a good one at that).

Now it's time to revisit this. I had forgotten there is so much to it.

Anyone want to make me an action that starts with a large original file, pauses for masking or whatever, and ends with a sharp teeny low-rez image? Oh, and I'll want the instructions to go with it, too

HAHahahaha

....well, I can dream, right?

PS: here are two links....Original shots, and Doctored ones...in case this helps anyone help me.

ORIGINALS
http://drcarl.smugmug.com/gallery/5542194_vQmHK/1/339464776_YFGpP

DOCTORED
http://drcarl.smugmug.com/gallery/5542214_h8z2A/1/339467289_fCWkn

Iceway

try this software "Image Optimizer"
"Image Optimizer" is a very usefull software in optimize pictures,
and you can try to use it with the "magic reduce" feature .
It can reduce the zise of picture to 10%~20% of itself, and what
is important, the picture quality almost no decline.