Does perspective control in PS degrade image quality?

toadlet

Hello

I was curious to know if the perespcitve fix on Photoshop when using the crop tool degrades image quality?

From the looks of it I would generally say no, but there have been a few images where I have made the buildings straight and on close 100 per cent inspection it seems a little more mushy/soft.

arodney

toadlet wrote:

Hello

I was curious to know if the perespcitve fix on Photoshop when using the crop tool degrades image quality?

From the looks of it I would generally say no, but there have been a few images where I have made the buildings straight and on close 100 per cent inspection it seems a little more mushy/soft.

Visually, maybe no. But there is data loss. But that's the case whenever you alter the RGB numbers from one set of values to another. If you need a perspective crop (or a burn and dodge, or level etc), and that improves the image appearance, the data loss is the worthwhile compromise. If you do all the work in 16-bit, the data lost is moot anyway (you have plenty of bits for output).

Duffy Pratt

I would think that degradation is mainly a subjective term. If your perspective change improves the image, then in what sense do you mean degradation.

Also, the notion that every RGB move causes data loss might depend on your definition of data. Suppose a given pixel in an sRGB jpg has the numbers 176, 176, 176. You make a curve in the R channel only, and now the pixel has the numbers 155, 176, 176. Was data lost? Well, you no longer have the original data, so in one sense yes. But on some other views of data loss, maybe not.

Now take this example. The entire image is gray - 176, 176, 176. You apply a gradient to the image, and the result is a gradient from white to black. Is there data loss or data gain? From the standpoint of preserving the original data, there is certainly data loss -- only a very small portion of the gradient will retain the values 176, 176, 176. But from another standpoint, the gradient certainly contains more discreet values than the gray field, so there is data gain.

Usually, when people talk about data loss with images, they point to holes and spikes in histograms, and show that the image, after a move, has fewer discreet values (colors) than it had before the move. It's possible, however, to perform moves that either maintain the number of values, or even increase them. That doesn't mean that the added "data" is a good thing (or a bad one).

Duffy

BinaryFx

I personally would phrase the question as "does such an image transform use resampling?" and the answer would be "yes". Although bit depth can have an effect on the interpolated data, one would be running into problems with either the resampling method and or the transform paramaters for the image content/resolution before the bit depth would cause noticeable issues. So for me, I would prefer to have excess pixels and only 8 bpc of data than only the required amount of pixels in 16 bpc which are then being resampled at 16 bpc. For many edits, having excess pixels and resampling the final output image down has a more noticable positive effect on visual image quality than bit depth does. In an ideal situation, one would have both excess pixels for transformations and high bit as well.


Regards,

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

arodney

BinaryFx wrote:

So for me, I would prefer to have excess pixels and only 8 bpc of data than only the required amount of pixels in 16 bpc which are then being resampled at 16 bpc.
http://members.ozemail.com.au/~binaryfx/

So would I if I knew exactly what excess pixels where. I've got a 12 megapixel capture device. If I knew I had to do perspective crop and could switch to a 22mp, OK that' be nice (although most of us what all the data possible for any eventual output size, hence the confusion about what excess pixels are).

I'd prefer excess pixels in high bit given a choice. OR better, a PC lens!