Sharpen AI of B&W images

I installed the trial version of this to hopefully replace the DxO NIK sharpener (which is rife with halos), but ran into this problem.

After running it on a black and white image, I found that the tool added a bunch of color information. See the below/attached image. Two rows (2 samples), original on left, sharpened with color on right.

The workflow started with a color image, which I processed with DxO NIK SilverFX Pro, then sharpened.

I’ve always thought of sharpening as a last step / output image after all other processing. Do I need to be doing the sharpening step to the original image before processing (or at least before B&W conversion)?

[ IMAGE: posting to these forums is frustrating. It’s telling me that I can only post images of specific types, but when I do, it tells me that I cannot put images in a post. (it is a reduced size). Is this yet another badge I have yet to earn? ]

1 Like

You have now been allowed to post images now, please upload so we can look at it.

Also, if you are using PS as a base make sure you use a stamped layer to pass to Sharpen AI to negate any making etc.

Thank you, image below.

Description: Two rows (separate samples), original B&W on left, sharpened on right where you can see the added color artifacts.

Note: so far, I’ve fed nothing but clean PS layers to SharpenAI (no stamping required).

Hi, can you please raise a support request at the main website as I haven’t seen this before …


1 Like

If you open an 8 bit or 16 bit greyscale D50 image in Sharpen AI freestanding, and export the sharpened image as a 16 bit TIFF with the color profile left as “Preserve source profile” (there are no greyscale options) , you will find that the exported image is 16 bit RGB, so perhaps it is not surprising that some of the pixels are no longer true shades of grey.

As you are starting with a color image in the first place and using Sharpen AI as a plugin, I would follow the Sharpen AI process with a (or another) convert to Greyscale. That conversion does not affect the sharpening, only the color.

He is starting with a RGB B&W image (TIF) produced from DxO NIK SilverFX Pro, not a Greyscale. It is totally unnecessary to use Greyscale for B&W images as you will lose the ‘color information’ held in an image.

A RGB Image is a three dimension (channel) image. It contains the Red color , Green color and Blue color image in separate matrix. …A GrayScale Image is a one dimension (channel) image and it is derived from a RGB image.

I’m well aware of that. Perhaps I should have said “Even if you start with a Greyscale…”. The point I was making is that Sharpen AI is exporting RGB images that have colors that are not shades of grey. Either way a convert/reconvert after the Sharpen AI should produce a suitable result. :grinning:

Convert to what, it is already B&W and shouldn’t be affected by Sharpen AI.

I’m not suggesting that Sharpen AI should be producing the results it is producing. I can produce similar results to what the OP has. I don’t know whether it is a bug, a shortcoming, or something unavoidable with the AI process.

At the moment it is what it is. But if the B&W conversion is not done until after the Sharpen AI process (which was raised by the OP), or redone after, even as a simple B&W filter , all the pixels with then be shades of grey (the numeric values for R, G and B will the same). So call it a work around if you like.

Thanks all.

It is indeed a workaround. And in my work-flow, it is a reasonable one for now.

I did a series of tests with combinations of SharpenAI, ACR, with NIK SilverFX Pro (my most typical means of going B&W with an image, but I use others and I’m guessing they are similar). I did these steps in various orders to see if there was a cost in image quality for this workaround.

I found that doing sharpening before and B&W processing after gave perfectly fine results with this relatively simple workflow. I think it will work fine for now with most of my work. This is all with RGB-mode 16-bit images.

However, that workflow (ACR, SharpenAI, B&W conversion) is relatively simple. I worry about images that require more sophisticated PS techniques with many layers and masks, grain and blur management, working with processes outside of PS that work with intermediate files. As I think I mentioned at the top of this thread, I would tend to think of sharpening as a final step in those kinds of situations.

But I still can do sharpening later for color output images. So far, for me personally, my B&W work doesn’t get that complicated. And if it does, I can still leave the B&W conversion as a later step, even if sharpening happens after all the PS voo-doo.

Also, one can add a simple Hue/Saturation adjustment layer on top and de-saturate all pixels there. That’s a small step.

As an aside, I find it interesting that the images still appear B&W when not pixel-peeping. :wink:

Cheers, Chris

I have tried the same where it is ACR, Silver Efex & Sharpen but haven’t come across the same issue as you have with colored pixels appearing in the output image from Sharpen AI … BUT … I am using the original Silver Efex Pro not the DxO version.

So, hopefully, support will be able to answer the question as to why colored pixels reappear.

I don’t think DxO has made real significant changes to the tools since acquisition from NIK. In fact, their more recent versions seem to be nothing more than more presets and some UI changes, no real advancements to the tools.

To update this community thread from the Topaz Support ticket I opened, here is their response:

“Sharpen AI is not yet trained in black and white images so sometimes this causes a conflict for the AI model. Not always, but it’s possible. The AI model is only trained with color so you’ll wnat to run a color image in Sharpen AI and convert to black and white afterward. We hope to add dedicated black and white AI models in the future though.”

Just FYI,

1 Like