View Full Version : Converting to another image size
CGNKlaus
21st of June 2004 (Mon), 08:54
Hi and regards from Cologne, Germany !
I just bought a 300D and went to a short trip to Paris. So I got some nice pictures.
One question:
I convert them with Xnview or Irfanview to a smaller size. Can you give me some tips to get a better quality ? I switched f.e. to 90% JPEG-quality, but I'm still not satisfied with it.
scottbergerphoto
21st of June 2004 (Mon), 09:10
You shouldn't lose any image quality when reducing the size of a picture. Make sure you don't have resampling checked for making smaller pictures.
Scott
Scottes
21st of June 2004 (Mon), 09:22
Actually, you do want resample selected. Resampling will average new pixels from the original pixels, whereas a simple resize generally throws pixels out. That is, if you're converting an image from 3000x2000 to 1/4 the size, 1500x1000, then Resample will take 4 pixels and average them, getting the color of the new pixels. Resize will pick 1 of those 4 pixels and that will be the color of the new pixel. (Simplified process...)
Using IrfanView I choose Image... Resize and shrunk a couple images down to either 1/2 or 1/4 the size. I used the default... Resample, using Lanczos. I was pretty happy with the results, considering that it was just a resize - the new image needed a bit of sharpening which is to be expected.
So I tried IrfanView's sharpen and it wasn't too bad at all. It wasn't perfect, but it was pretty good for a 1-step operation with no options.
But doesn't Photoshop Elements come with a 300D? That will do a much better job of resizing and sharpening - at the expense of more work on your part.
scottbergerphoto
21st of June 2004 (Mon), 11:15
Actually, you do want resample selected. Resampling will average new pixels from the original pixels, whereas a simple resize generally throws pixels out. That is, if you're converting an image from 3000x2000 to 1/4 the size, 1500x1000, then Resample will take 4 pixels and average them, getting the color of the new pixels. Resize will pick 1 of those 4 pixels and that will be the color of the new pixel. (Simplified process...)
Using IrfanView I choose Image... Resize and shrunk a couple images down to either 1/2 or 1/4 the size. I used the default... Resample, using Lanczos. I was pretty happy with the results, considering that it was just a resize - the new image needed a bit of sharpening which is to be expected.
So I tried IrfanView's sharpen and it wasn't too bad at all. It wasn't perfect, but it was pretty good for a 1-step operation with no options.
But doesn't Photoshop Elements come with a 300D? That will do a much better job of resizing and sharpening - at the expense of more work on your part.
Not wanting to get into a pissing match, but needing to make the point, when downsizing digital images Scott Kelby in The Photoshop CS Book for Digital Photographers, page 73, "......turn the Resample Image checkbox off(it's on by default).""Resampling while downsizing makes an image look blurry and pixelated".
The advice for increasing the size of digital images is the opposite. On page 75, of the same book, "......make sure Image Resampling is turned on."
Scott
Scottes
21st of June 2004 (Mon), 11:56
When you work in Photoshop and run Image... Image Size and uncheck Resample Image then you're not resizing the image, you're simply changing the DPI. Once you uncheck Resample you disable the text boxes used for dimensions. So you're not changing the size, you're changing the resolution. This means that you'll change the size of the print but not the size of the image.
Now bring up an image in either Irfanview or Xnview - which are the programs CGNKlaus mentioned. Resize an image using both Resize and Resample. Sharpen them. Which looks better? I used Irfanview, and Resampling using Lanczos looked much better to me. To me, using Resize on my sample image made it look like crap. I haven't used XnView in some time, but I'm sure that the results will look similar.
drisley
21st of June 2004 (Mon), 12:00
Lanczos interpolation is one of the most advanced resize algorithms around, much more advanced than bicubic. I still dont know why Photoshop doesnt start using Lanczos.
For downsizing, I find the best method is to decrease in 50% intervals, then when you are 1 size bigger than your final size, decrease to the size you want. Then add a little sharpening.
Ive never tried downsizing without resampling. I might have to give it a try. Photoshop doesnt actually give you this option (you only change the dpi this way).
scottbergerphoto
21st of June 2004 (Mon), 12:13
When you work in Photoshop and run Image... Image Size and uncheck Resample Image then you're not resizing the image, you're simply changing the DPI. Once you uncheck Resample you disable the text boxes used for dimensions. So you're not changing the size, you're changing the resolution. This means that you'll change the size of the print but not the size of the image.
Now bring up an image in either Irfanview or Xnview - which are the programs CGNKlaus mentioned. Resize an image using both Resize and Resample. Sharpen them. Which looks better? I used Irfanview, and Resampling using Lanczos looked much better to me. To me, using Resize on my sample image made it look like crap. I haven't used XnView in some time, but I'm sure that the results will look similar.
An image has a finite amount of data, expressed as ppi. If you take a large image and make it smaller without tossing out pixels, the ppi goes up. So yes, your increasing resolution, but not by adding pixels as you do in upsizing. You do it by moving the pixels closer together to get a smaller image. I don't know why you say that isn't resizing. Scott Kelby obviously calls it that.
Scott
Scottes
21st of June 2004 (Mon), 12:41
An image has a finite amount of data, expressed as ppi. If you take a large image and make it smaller without tossing out pixels, the ppi goes up. So yes, your increasing resolution, but not by adding pixels as you do in upsizing. You do it by moving the pixels closer together to get a smaller image. I don't know why you say that isn't resizing. Scott Kelby obviously calls it that.
But increasing or decreasing PPI/DPI you are not changing the size of the image, you are changing the size of the print. DPI/PPI is meaningless on a computer screen - changing it will not change the size of the image as displayed on the screen. Whether the image is 2000ppi or 72 ppi, both XnView and Irfanview will display it identically.
The term "Resizing" has no inherent direction - it could be upsizing or downsizing.
To clarify "throwing out pixels" I should have said "ignoring them." Resampling algorithms such as bicubic or Lanczos or weighted-average (and so on) will take a block of pixels and - in some way - average them to get the value of a new pixel. This new pixel might be added to the image (when upsizing) or be used as the resultant pixel in the new image (when downsizing). A Resizing algorithm, on the other hand, generally just picks on the pixels as the resultant pixels. When downsizing to 1/2 size, a resizing algorithm simply picks every other pixel to be the resultant pixel.
And Scott Kelby's term is correct. However, it's only correct when talking about the final destination being a print. Again, PPI/DPI is meaningless when talking about images displayed on a computer screen.
And I'm pretty damn sure that I'm not disagreeing with him. My previous posts use the term Resizing to indicate the algorithm used in Irfanview and XnView. Resizing as opposed to Resampling - those are the two options in irfanview, and are generally regarded as the two possible ways to change an image's size. Within those two ways there are a number of different algorithms - though Resize has only one because if it really used a different algorithm then it would be Resampling.
vBulletin® v3.6.12, Copyright ©2000-2012, Jelsoft Enterprises Ltd.