Adobe shows off Content-Aware Removal for Photoshop

[neshcom]

http://www.labnol.org/software/removing-ob...hotoshop/10675/



"October 20, 2009

Removing Unwanted Objects from a Picture in Photoshop Could Become Child’s Play



Adobe Photoshop is often used for removing backgrounds and other unwanted objects from a photograph but this new "content aware" technology not only improves that process but makes it easy for Photoshop newbies as well.



You just have to paint the object that you want to remove using the Spot Healing Brush and the program will automatically fill that area with a surrounding pattern.



While Adobe is not talking dates, we could see this feature live in Photoshop CS5."



<img src='http://mm-bbs.org/public/style_emoticon ... >/w00t.png' class='bbc_emoticon' alt=':)' /> This is really cool stuff. No longer do you have to deal with terrible, terrible spot healing, especially on larger objects. :mouthagapeforever:

[Zaeleus]

I recently read a paper concerning patched-based synthesis (what this article is talking about). The most mind-boggling part is that the techniques used are randomized algorithms, that is, the program must sample random parts of the image hoping it'll find a patch that is close to what it needs. Despite thinking that this might be the most inefficient way of handling the situation, the improved performance gained by these algorithms allows it to run at near real-time speeds compared to the much slower patched-based methods used in the past, making it suitable for interactive editing like in Photoshop. If anyone is interested, the paper can be found on Scribd (Adobe Photoshop CS5 - PatchMatch - www.cs5.org).

[aine]

This is very cool but not new at all, exactly the same function has been in at least Capture NX2 for at least a year now. And does work as simple as that - you mark the unwanted object with a brush, and the software removes it based on similar patterns in the image. Very often with surprisingly good results, much better than you could get with just a clone brush. Although I guess that's good news for all Photoshop users.

[surasshu]

This reminds me of that content-aware resizing thing. Whatever happened to that?

[neshcom]

[quote name='///' post='70459' date='Oct 21 2009, 03:47 AM']This reminds me of that content-aware resizing thing. Whatever happened to that?[/quote]

I've never seen the constant window resizing, but I know you can you use content-aware resizing in PSCS4.

[Ap2000]

I SO doubt that is going to actually work very well.

Especially when you're doing something with a really high-resolution picture, you probably will still have to retouch a lot.

[Zaeleus]

[quote name='aine' post='70458' date='Oct 21 2009, 02:07 AM']This is very cool but not new at all, exactly the same function has been in at least Capture NX2 for at least a year now.[/quote]

Not exactly. I downloaded the Capture NX2 trial to test the auto retouch tool, and it makes some very similar mistakes compared to other proximity based patching tools. It is very much unaware of nearby patterns. Yes, the method in which the user interacts with the image may be the same, but the paper presents a completely different implementation.

[quote name='///' post='70459' date='Oct 21 2009, 02:47 AM']This reminds me of that content-aware resizing thing. Whatever happened to that?[/quote]

Seam carving tools have been around since the CAIR paper was published (e.g., online tool rsizr, GIMP plugin Liquid Rescale, Photoshop's content aware scaling, etc.), but it doesn't seem to be widely used. The problem with seam carving is that it tends to provide unsatisfactory results, as the energy function does not compensate well for the way we perceive a particular image. Parallel lines, either simple parallel ones or ones that converge to a vanishing point, cannot easily be weighted using a soft brush, so they become warped in the process. An alternative seam carving algorithm is described in the PatchMatch paper, and it can manage and propagate repeated patterns in an image, such as these repetitive lines. When the algorithm fails to find suitable patterns to preserve the linear parts of an image, it's possible for a user to provide constraints on the lines (similar to interactive weighting but provides a more defined edge).

[fpd]

I'm most excited about the new Flash, but the stuff I've seen from photoshop looks great.