The Case For JPEG; It’s Not Just A Small Print And E-Mail Format Page 2

sorcadmin's picture
Using Subsampling To Your Advantage
I don't mean to sound as though I'm against subsampling. Used correctly, it's actually a helpful feature. Keep in mind that when you shoot a photo, you have no definitive way of knowing whether subsampling will affect the photo's visual quality. However, when you're preparing images in Photoshop, you have the advantage of Photoshop's side-by-side previews to determine if a given level of compression/subsampling is affecting an image's quality. In a case such as this, you may be able to apply subsampling, reduce file size even more, and do so with the photo's visual quality completely untouched. In a situation such as this, subsampling now becomes a useful feature.

No Subsampling vs. Subsampling. Image 4 (above) was saved on a setting of 90 (on a scale of 0-100) with Subsampling turned off. Image 5 (below) was also saved on a setting of 90, but with Subsampling turned on. The difference may not be obvious to the naked eye, but Images 6 and 7 will show that the damage is still there.

It may surprise you to learn that, just as hardware vendors are secretive about subsampling, so are many software vendors. Over the years, Adobe and Macromedia have both refused to discuss when subsampling is applied. They'll only echo what is said by hardware manufacturers: At some point, subsampling is turned on, and at some point, subsampling is turned off.

If you're interested in fine-tuning your JPEGs with the perfect combination of subsampling and quality, there are a small handful of software vendors that do provide direct control over subsampling. Corel's PHOTO-PAINT (the photo editor in Corel's Graphics Suite) enables two sampling options: "444," which is the equivalent of no subsampling, and "422," which applies mild subsampling. Ulead's PhotoImpact also makes the 444 and 422 options available, along with 411, a stronger form of subsampling.

I've found that the best way to use subsampling is to first decide on a quality setting with subsampling turned off. Once this is done, try 422 or 411 subsampling. In many cases, this will cause absolutely no visual change in your image (check the side-by-side previews to be sure), and will result in a smaller file size.

These images are crop-outs of Images 4 and 5, respectively. Although zooming in on an image always causes some damage, the difference in these images will be clear. Image 6 (above) shows clean-cut lines with only minor zoom blurring where abrupt color changes occur. In addition to the zoom blurring, Image 7 (below) shows clear JPEG artifacts along the edges of each of the white and green color blocks. These artifacts are the result of the JPEG subsampler averaging the color values of the black lines with the white or green color blocks, and coming up with color values that don't match the original values. As your eyes move away from the black lines, you'll notice the black lines have less and less influence on the color averaging, and eventually the green and white color blocks return to their normal values. You could print Images 4 and 5, or post them on a website, and you may never see a problem. The risk comes in using them for long-term storage. If, at some point down the road, you wanted to do some additional post-processing on a subsampled JPEG, these artifacts may well become visible. The bottom line here is, for that one original digital image you plan on keeping long term, as a replacement for a film negative, or a replacement for a TIFF, you should always use non-subsampled JPEG to ensure a high-quality original.

In Conclusion
The bottom line with JPEG is that if you avoid the pitfalls (using it on computer-generated images, using subsampling on originals, and adding subsampling/extra compression without using side-by-side previews), there is absolutely no visual difference in a properly saved JPEG, and in a much larger file such as a TIFF, PSD, or PNG.

Anthony L. Celeste appreciates feedback from his readers. You may contact him via e-mail at

Article Contents
Share | |