Requested Improve the Algorithm for Downsizing Images

[LaMo]

Hi Irfan,

I appreciate the application Irfanview very much!

One detail though:

I think I get a better result doing manual transfer to jpg than your application.
Maybe you should consider your algoritm to downsize to jpg?

Attached are two images ’byMe.jpg’ and ‘byIrfan.jpg’.

What bothers me about the Irfan image is the jagging which is visible on the tram in the lower right corner.
Any sloped line gets a disturbing jagging.

I do the downsizing (using a .psd-file, which in the last step is saved as a jpg-image) manually in Photoshop (CS4) following this routin:
The image starts at a size of 650 dpi (175 MB).
I resize it to 600, 550, 500, 450, 400, 350, 300, 260, 220, 180, 140, 110, 90 and 72 dpi.
At about every third resizing I apply sharpening with Unsharp Mask at pixelsize 0,2 and 140-500%.

Now when you compare the two images you can see that the Irfan image has a slightly better detail sharpness than the byMe image.
On the other hand the jagging is softer on the byMe image.

The better detail sharpness in the Irfan image is only visible at a big enlargement not displayed in 100%. The byMe image also has a better final definition in 100% because of the sharpenings.

Of these two images I would prefer to use the byMe image on my web, because of no hard jagging, and a better sharpening definition.

My routine is to use Irfanview when downsizing. When I see the visible jagging I redo that image manually.

Of course my test is only applies to downsizing ca 150 MB files to 24cm/72dpi web images. I really can’t say whether the jagging also appears using other sizes.

If you agree with me maybe it is possible to adjust your downsizing algoritm?

Regards

Lars Mongs

[derniwi]

Hi,

there is no perfect algorithm for downsizing images. So if a small image does not look as expected you have to try different ways. Also using PS and unsharp masks is not the best way for sharpening images...

If you search the web you may find other methods for resizing which will give good results on specific pictures.

So I think sometimes it is good to share ideas and methods how things can be done. But I don't believe that Irfan is going to implement a lot of algorithms for downsizing images. :-)

Best regards,
Nils.

[LaMo]

Hi Nils,

Thanks for responding.

I don’t follow you completely.

I thought that Irfan mainly just worked with converting and resizing algoritms, but you may be right, they maybe are not interested.

My use of PS and unsharp Mask at least beats Irfan’s jpg algoritm.
So it cannot be that bad.

After your suggestion I googled on the internet and tried some jpg-converters that came up.
Tried to resize and convert a 175 MB image to a small (600kb) web-jpg:

Silver Eagle Software – horrible and useless result.
ReaConverter – not possible to understand how to resize (mentioned in their so called Help, but not possible to actually do in their interface).
Pixillion – can convert but cannot resize.
AVSImage converter – horrible result.

So those are pretty much useless.

Now I wonder if you have heard of any specific application that is good at downsizing and converting to jpg (to make web images out of high resolution images)?

Thanks for taking interest in my problems…

/LM

[derniwi]

Hi,

I agree, your algorithm is not bad. About sharpening you may read or still know this article:

http://ronbigelow.com/articles/sharpen1/sharpen1.htm

In the last years I do just one sharpening and resizing step, which is good enough for my pictures:
1. save the picture
2. save the picture with a new file name
3. reduce the picture to one layer (background)
4. resize
5. duplicate the layer
6. reduce the saturation of the new layer, so it became gray scaled
7. set the layer mode to hard light
8. use the high pass filter with a radius from 0,5 - 10 on the new layer (you can test and see the effect in 100% zoom)
9. if the effect is to strong you may change the opacity for the high pass layer or use a layer mask to reduce the effect

But as I said before, there is now universal sharpening algorithm for every purpose.

Regards, Nils.

[LaMo]

Hi Nils,

Thanks for the article. I will read it with great interest.

I also do high-pass-filter sharpening, but on the full resolution pictures.
New to me was to take down the saturation on the duplicate layer.

Very nice, I have some testing to do!

Thanks again!

/LM

[derniwi]

Sharpening is almost changing the contrast on edges, so if you use a high pass filter and keep the layer colored, it will not only change the contrast. For this it is necessary to remove the colors from the sharpening layer.

[LaMo]

WHen using the high pass filter I also find it useful to cut out most of the highlight parts in the duplicate layer (done in Levels). I only want darker parts to be enhanced.
So maybe that was some kind of color reduction too.