Resizing quality and 8/16 bit.

[Aesthete18]

Hi I was just wondering a couple of things:

1. I open a huge picture>ctrl+T>resize it to about 90% smaller and I can tell the quality is not as good. Should it be better when it's smaller? I don't understand it.

2. I open a picture that's 8 bit (why is a picture 8 bit vs 16 bit?) then I resize it and what not and drag it into the main project and it says "the target document has a different depth.." How much of the quality does it affect? I shrink most of the pictures about 90%. Also, is there a way to make the original pictures 16 bit instead of 8 bit?

Sorry about the terms and such, I'm not that familiar with the Photoshop lingo.

 

[Tom Mann]

1. The most common reason people experience this loss of sharpness is because after they down-rez their image, they then increase the magnification to make the down-rez'ed version of the image about the same size on their screen as the high rez version was before down-rez'ing. Of course, since one has discarded many pixels in the down rez'ing process, it contains less information and looks worse.

Another way to think about this perceived loss of sharpness is the size of each pixel vs the size of the smallest features visible in the image. Suppose you start out with a nice sharp head-and-shoulders portrait in which each tiny hair on the face is around 1 pixel wide. If you down-rez this image by 10x (linear dimensions), each pixel will now be 10 "hair widths" (or, say, 1 zit wide wide, LOL) instead of 1 "hair width" wide, so, of course the down-rez'ed version looks less sharp.

2. The bit depth of an image has nothing to do with the spatial resolution. An "8 bit" image means that the brightness of each of the three color channels (ie, R, G, and B) is represented by an integer that can only run from 1 to 256. Put differently, such an image has only 256 different levels of red, 256 different levels of green, etc. etc.

In contrast, in a "16 bit" image the brightness of each channel is represented by an integer that can run from 1 to about 65,000.

Visually obvious problems can easily arise with 8 bpc images. These problems include banding in slowly and smoothly varying areas such as clear skies, a "crunchy" look or noise when you try to brighten deep shadows in an image, and lack of smoothly varying or even any detail in very bright regions (ie, because, say, the red channel in your flower photo is maxed out at 256 in certain parts of the flower). 16 bpc images were developed to minimize such problems, and works incredibly well.

Bit depth issues pertain to each pixel separately. They have next to nothing to do with spatial resolution, ie, how many pixels wide should the image be to accurately represent the tiny hairs on someone's face.

As a side comment, I use 16 bpc images for probably 95% of my work in Photoshop. About the only time I'll ever go back to 8 bpc images is if I am forced to because I want to use some old plugin, software, or data format that only supports 8bpc images. The prime example of this are JPG files. They are all 8 bpc.

To convert an image between 8 and 16 bpc, just select the bit depth that you want in the "image / mode" pull down menu.

Finally, I should comment that the bit depth of 99.99% of all images are either 8, 16, or 32 bits per channel. However, operating systems also come in "32 bit" and "64 bit" versions. Because the term, "32 bit", occurs in both places, people sometimes confuse the two numbers. In contrast to the bit depth for images, the numbers for OS's tell you about the maximum amount of physical memory in your computer that can be utilized, the size (in bits) of each chunk of data/instruction sent to the CPU, etc.

To distinguish between the two usages, I usually write "bpc" after the number when it pertains to images, and "bit OS" when it pertains to operating systems. So, one can (in theory, at least) have 8, 16, or 32bpc images being processed in either 32 or 64 bit OS's.

HTH,

Tom M

 

[Aesthete18]

1. The most common reason people experience this loss of sharpness is because after they down-rez their image, they then increase the magnification to make the down-rez'ed version of the image about the same size on their screen as the high rez version was before down-rez'ing. Of course, since one has discarded many pixels in the down rez'ing process, it contains less information and looks worse.

Another way to think about this perceived loss of sharpness is the size of each pixel vs the size of the smallest features visible in the image. Suppose you start out with a nice sharp head-and-shoulders portrait in which each tiny hair on the face is around 1 pixel wide. If you down-rez this image by 10x (linear dimensions), each pixel will now be 10 "hair widths" (or, say, 1 zit wide wide, LOL) instead of 1 "hair width" wide, so, of course the down-rez'ed version looks less sharp.

2. The bit depth of an image has nothing to do with the spatial resolution. An "8 bit" image means that the brightness of each of the three color channels (ie, R, G, and B) is represented by an integer that can only run from 1 to 256. Put differently, such an image has only 256 different levels of red, 256 different levels of green, etc. etc.

In contrast, in a "16 bit" image the brightness of each channel is represented by an integer that can run from 1 to about 65,000.

Visually obvious problems can easily arise with 8 bpc images. These problems include banding in slowly and smoothly varying areas such as clear skies, a "crunchy" look or noise when you try to brighten deep shadows in an image, and lack of smoothly varying or even any detail in very bright regions (ie, because, say, the red channel in your flower photo is maxed out at 256 in certain parts of the flower). 16 bpc images were developed to minimize such problems, and works incredibly well.

Bit depth issues pertain to each pixel separately. They have next to nothing to do with spatial resolution, ie, how many pixels wide should the image be to accurately represent the tiny hairs on someone's face.

As a side comment, I use 16 bpc images for probably 95% of my work in Photoshop. About the only time I'll ever go back to 8 bpc images is if I am forced to because I want to use some old plugin, software, or data format that only supports 8bpc images. The prime example of this are JPG files. They are all 8 bpc.

To convert an image between 8 and 16 bpc, just select the bit depth that you want in the "image / mode" pull down menu.

Finally, I should comment that the bit depth of 99.99% of all images are either 8, 16, or 32 bits per channel. However, operating systems also come in "32 bit" and "64 bit" versions. Because the term, "32 bit", occurs in both places, people sometimes confuse the two numbers. In contrast to the bit depth for images, the numbers for OS's tell you about the maximum amount of physical memory in your computer that can be utilized, the size (in bits) of each chunk of data/instruction sent to the CPU, etc.

To distinguish between the two usages, I usually write "bpc" after the number when it pertains to images, and "bit OS" when it pertains to operating systems. So, one can (in theory, at least) have 8, 16, or 32bpc images being processed in either 32 or 64 bit OS's.

HTH,

Tom M

Very well explained. A lot of the pictures I had to use were in 8-bit so it's good to know I can convert them.

Of the five pictures I had to down size yesterday, two seemed to maintain the quality or at least not be as affected. Also, I didnt zoom in or anything. Perhaps my eyes isn't trained enough to see the change in the quality yet. I thought I was resizing the wrong way but seeing that you didn't offer a solution, I'm guessing there isn't one.

Thank you for the long and thorough explanation and without using many terms, it is appreciated.

 

[Pixel8r]

Best way to see difference is use gray color and gradient tool. Change between 8 and 16 bit. You can see how much smoother 16 bit is. Notice that in 16 bit mode there is fewer options in filters. I usually change to 16 bit after everything is 99,9 % ready.

 

[Tom Mann]

Pixel8r: "...I usually change to 16 bit after everything is 99,9 % ready...."

I presume that was a typo and you meant to say the opposite.

One gains nothing by switching to 16 bpc after all the processing of the image is complete. The advantages of working in 16 bit mode (ie, over 8 bpc mode) occur when you are doing operations that change the numbers in each pixel. Each such operation is essentially a mathematical operation, and one builds up rounding errors with each such operation. Each of these errors will be about 256 times smaller if you work in 16 bpc vs 8 bpc.

Tom

 

[Pixel8r]

One gains nothing by switching to 16 bpc after all the processing of the image is complete.

Well, that depends. If you have a let say gradient background (w filtered layers) which looks compressed in 8 bit mode, when you switch it to 16 bit mode it becomes smoother. But of course if you like to get all advantage from 16 bit accuracy, stay in that mode from the beginning. Of course it takes some speed from your workflow and file size is bigger.

 

[Tom Mann]

"Well, that depends. If you have a let say gradient background (w filtered layers) which looks compressed in 8 bit mode, when you switch it to 16 bit mode it becomes smoother."

I'm sorry, but no it doesn't. Think of it this way.

Suppose you have an 8 bpc B&W image with a smooth, small gradient in brightness and you see banding. You move your eyedropper tool over that area and you read out integer values like 200, 201, 202, etc. Each numerical step corresponds to one visually apparent band.

If you now convert that image to 16 bpc, the numerical values will now be (converted to decimal, but still on the same 1-256 scale) 200.000, 201.000, 202.000, etc. Conversion to 16 bpc does not, in itself, produce any new values that are in-between the old numbers, it merely adds more precision to each of the old numbers, so, you still have banding.

Now, of course, once you are in 16 bpc mode, you can do things like intentionally add a bit of noise to randomize the numbers in each of the old bands. For example, all pixels throughout the 2nd band formerly had one single value (201.000). However, once you add a bit of noise, the pixels in that band might have now values from 200.500 to 201.500 thereby somewhat breaking up the sharp edges between the bands, but this is a step you have to consciously take. It doesn't come from just the act of changing the bit depth.

T

PS - BTW, this subject has been hashed out countless times in previous discussions on the net, particularly, in forums such as on photo.net. If you would like, I'll try to find references to some of those earlier discussions for you, but there essentially is no debate on this any more: stay in 16 bpc mode for as long as possible, and convert to 8 bpc only when you absolutely must (eg, final production of an 8 bpc JPG for posting to the web).

PPS - BTW, if you do the "add noise" routine as described above to try to break up banding, drop the spatial resolution a bit (or squint, LOL), and the banding becomes apparent again.