Book advice on cloning techniques

mercphoto

I'm looking for a book on Photoshop that would delve into cloning techniques, especially on recreating pieces of an image that might be obscured by a foreground element. Clone out the foreground element and recreate what was previously hidden from view. I'm thinking one of these two books but certainly open to suggestions:

"Adobe Photoshop Restoration & Retouching"
http://www.amazon.com/Photoshop-Restoration-Retouching-Voices-Matter/dp/0321316274/ref=wl_it_dp?ie=UTF8&coliid=IG0RIS0XWY3DQ&colid=1WMMQQUY0SU6Y

"How to Cheat in Photoshop CS3"
http://www.amazon.com/How-Cheat-Photoshop-CS3-photorealistic/dp/0240520629/ref=wl_it_dp?ie=UTF8&coliid=I2S4TEYVJN78F&colid=1WMMQQUY0SU6Y

Thanks!

pathfinder

Like how to remove this wire? I am all ears



With only one frame I think it would be tough to look like it had not been obviously edited.

But if one were to shoot 3 to 6 frames, varying the angle of the camera slightly to shift the registration of the wire in relationship to the lion, the images could then be stacked in the Layers palette, and the various portions of the wire masked out. This should work fairly well with a stationary subject a little removed from the wire fencing. Doing this with a living, breathing, moving lion may be challenging.

With only a single frame, you might make several copies of it and place them in the layers palette slightly out of register to allow masking out the wire in various areas. Sometimes the area north of the wire, sometimes from the area south of the wire, or east or west. Using a single image like this keeps the color and tone accurate. I think this might be doable if one has the time and the patience.

mercphoto

pathfinder wrote:

Like how to remove this wire? I am all ears

Yup, like that. Another example is here. In this photo I cloned out parts of a rig that was attached to the hood of the car and suspending the camera in front and to the side. But I only cloned PART of the rig. Most of the rig I CROPPED out because the cloning was too hard. I would have had to reconstruct elements of the car, such as a hood seem, part of a light, etc.

pathfinder

For this you will need to shoot a frame or two of the car without the camera rig attached, but from the same actual camera position as when the camera is mounted in the rig. Then the two layer can be combined with masking in the Layer palette.

That make sense?

I think using multiple frames and masking them in the Layers palette will be easier and look more accurate than just trying to clone things in or out. The lighting will need to be the same as well, as the wheel direction too. You will only need one good frame of the car without the camera rig attached, but shot from precisely the same place as the camera sits when mounted in the rig. Then mask them together into one image.

mercphoto

pathfinder wrote:

For this you will need to shoot a frame or two of the car without the camera rig attached, but from the same actual camera position as when the camera is mounted in the rig. Then the two layer can be combined with masking in the Layer palette.

That make sense?

Yeah, makes sense. Not sure how easy it is to do in practice. By that I mean, getting the exact same camera location when the rig comes off the car. But I imagine if you can get it close that you can rotate and twist and re-size various tidbits to get it close enough? I should practice this and see what happens!

LiquidAir

mercphoto wrote:

Yeah, makes sense. Not sure how easy it is to do in practice. By that I mean, getting the exact same camera location when the rig comes off the car. But I imagine if you can get it close that you can rotate and twist and re-size various tidbits to get it close enough? I should practice this and see what happens!

Yep. Lay one on top of the other with the transparancy around 50% and then use the transform tools to line them up. Next, mask off the top layer so that it only covers the areas you want to replace. Set the transparancy back to 100% and use curves to match up the colors. I control-shift click on my top layer to set curves anchor points and then sample the lower layer to figure out what my target color is. Its a simple matter then of typing in the target R, G, and B values into the curves dialog (I wish Photoshop made this easier). In more difficult cases, I end up using multiple curves layers with masks to match the entire perimeter of the patch. Finally, you'll probably need to drop another layer on top and use the healing brush for cleanup.

mercphoto

LiquidAir wrote:

Yep. Lay one on top of the other with the transparancy around 50% and then use the transform tools to line them up...

I do remember two big problems I had. Remember, these shots that I'm interested in are motion shots with long shutters. When cloning on top the hood of the car, for example, two problems would manifest. One is the fact that any reflections in the hood are going to be motion-blurred, which will not happen in the static shot w/o the rigging. Two, even though the hood is a single color there is always a gradient and that made cloning difficult. Imagine a 1x1 aluminum tube over the hood and you need to clone that out, matching the color gradient one each side. Fortunately the blurry reflections are actually pretty easy to deal with. Its the color shift I found difficult to deal with. And looking back I'm wondering if the way to deal with that is a gradient mask on the cloned section itself?

LiquidAir

mercphoto wrote:

I do remember two big problems I had. Remember, these shots that I'm interested in are motion shots with long shutters. When cloning on top the hood of the car, for example, two problems would manifest. One is the fact that any reflections in the hood are going to be motion-blurred, which will not happen in the static shot w/o the rigging. Two, even though the hood is a single color there is always a gradient and that made cloning difficult. Imagine a 1x1 aluminum tube over the hood and you need to clone that out, matching the color gradient one each side. Fortunately the blurry reflections are actually pretty easy to deal with. Its the color shift I found difficult to deal with. And looking back I'm wondering if the way to deal with that is a gradient mask on the cloned section itself?

I have not tried anything as complicated as what you are doing, but the way I have dealt with these issues in the past is to match up one end of the gradient with a curves layer; add a second curves layer to match up the other end of the gradient; and then blend between the two of them by painting with a large soft brush in the layer mask.

jjbong

pathfinder wrote:

Like how to remove this wire? I am all ears

This is reminiscent of another thread [thread]64775[/thread] in which Fast Fourier Transform filters removed what appeared to be a similar problem - a net in front of a baseball scene.

This technique failed miserably here, at least for me. After I did the FFT, I didn't see the tell-tale periodic spots to wipe out, and my attempts to wipe out periodicity failed miserably, generating very blurry images.

Any idea why FFT doesn't work here?

LiquidAir

jjbong wrote:

This is reminiscent of another thread [thread]64775[/thread] in which Fast Fourier Transform filters removed what appeared to be a similar problem - a net in front of a baseball scene.

This technique failed miserably here, at least for me. After I did the FFT, I didn't see the tell-tale periodic spots to wipe out, and my attempts to wipe out periodicity failed miserably, generating very blurry images.

Any idea why FFT doesn't work here?

The fence is too coarse. When you do simple FFT interpolation like suggested in that thread, you are throwing away a range of the spectral information. With fine grids that spectral information is often not very important to the final image. However with a coarse grid like the fence it is going to be a problem. With a good mask for the fence and some fancier mathematics you can do better but there is just too much valuable information about that lion lost to the fence to get a satisfying image no how big a hammer you apply to the problem.

pathfinder

jjbong wrote:

This is reminiscent of another thread [thread]64775[/thread] in which Fast Fourier Transform filters removed what appeared to be a similar problem - a net in front of a baseball scene.

This technique failed miserably here, at least for me. After I did the FFT, I didn't see the tell-tale periodic spots to wipe out, and my attempts to wipe out periodicity failed miserably, generating very blurry images.

Any idea why FFT doesn't work here?

I remember that thread, but could not seem to find it - I will now bookmark it so I don't lose it again. I don;t think it will work on my fence though, it seems to work better on much smaller patterns.