Cardinal Band

High Quality
Check out this page if you are looking for High Quality

Creating high quality JPEG images

JPEG Compression - How does it work?

The JPEG file format is one of the most important innovations of the internet era. Its amazing ability to render high quality images from small files is achieved using the limitations in human vision to discard data you probably won't notice.

1. JPEG takes advantage of our eye's low sensitivity to details in certain areas by down-sampling (think of this like converting four pixels into one averaged pixel). However, it retains detail in areas of bright, high-contrast edges.
2. Instead of smoothing detail in the whole image, JPEG selects which regions of the image, or even just channels of colour in the image that can tolerate these averaging effects without being noticed. One example is that it uses our eye's reduced response to yellow-blue light (compared to red-green) and reduces the information stored in these colour channels.

In addition to these tricks on our perception of the image, JPEG compression removes redundancy - and that's one reason you can't compress JPEG files any further with traditional tools like ZIP.

How well does JPEG compression work?

Well, if we compare a typical JPEG file to a TIFF image - that's a reference format that traditionally uses no compression at all - then we find that JPEG files are generally as little as 10% of the file size of the TIFF.

The most striking benefit of JPEG is that a file that would take 10 seconds to load in your web browser now takes 1 second.

This is just scratching the surface though: what we must do is consider some of the downsides of JPEG compression, especially considering that graphics tools give you many choices when creating JPEG files. From there we can try and get the optimal file size while still obtaining a high quality level.

How to create high quality JPEGs

The first rule of high quality JPEGs is that you should never open, edit and save a JPEG image repeatedly.

If you do that, the losses in the compression method used by JPEG will accumulate each time you save and then reopen the file, so each successive version of the file will be slightly worse than the last version. Instead, use a lossless working format (covered in a forthcoming blog post).

Secondly: once a JPEG, always a JPEG. Don't convert a JPEG to another format like TIFF to recover quality, as you can't reverse the losses in the JPEG file.

The next big factor is the quality setting.

You will find a quality setting in any software that can save JPEG images. This is often a number from 0 to 100, or may be expressed as "high quality", "medium quality" and so on. In PhotoShop, the quality is graded on a scale of 1 to 12.

As you add more compression - ie. lower the quality setting - the file size shrinks but the image is visibly degraded more. This reduction in detail is a result of the lossy nature of JPEG compression.

The loss that you see will become much more apparent when the compression is set very high (ie. quality between 0 and 20), while correspondingly, as you increase quality, you increase the size of the output file far more rapidly as you approach 100.

Compression vs. file size curve

It's very important to know that the numbers we talk about with JPEG are not normally in a simple relationship - for example, choosing 95 on the quality setting does not mean there is a further 5% more quality left to give, and nor does a compression setting of 10 mean that the file can be reduced to ten times smaller by lowering the compression setting to 1.

These are both commonly held beliefs but quite wrong, and arguably with quality, impossible to quantify!

In order to help you choose an appropriate compression level, I've analysed a JPEG file containing some random content at different compression levels. Here is a graph showing the size of the file saved at each compression level from 1 to 100:

Important: the file size in bytes is on a logarithmic axis. The values double at each point on the file size axis.

1. You should be able to see that a high quality comes with an increasingly high storage price tag as you approach 100.
2. In fact, the largest file on the chart (quality 100) is about 50 times bigger than the smallest file.
3. At quality 95 it's only 20 times bigger. Setting the quality level just slightly lower significantly shrinks the file - and usually, for no perceptible change.

You are probably most interested in the high quality end (where all the real uses of JPEG are found). My advice is to keep the quality setting between 70 and 90 for good results. Above 95 is a very wasteful setting for all but the most specific end uses. Around 60 to 70 is perfect for good quality web use.

Diminishing returns on JPEG Compression

What about the lowest possible quality setting? We all know from experience with JPEG that the image will be very blocky and degraded. A much better result can be obtained for just a little more file size. How do we quantify this "quality" measurement though?

The nature of perception of quality makes it quite difficult to show you a graph of image quality versus file size. Remember, the graph above was about compression versus file size - it doesn't tell you how good the image looks after compression.

In the two examples below, I've zoomed in on a JPEG saved twice at different compression levels. When you look at these images, bear in mind that Image A is a mere 9KB larger on disk than Image B.

You can see more detail, ie. better quality in Image A. If you're in any doubt, look at the sky to the left of the nearest boat.

Chances are, you haven't accepted low quality JPEGs but you might well have been tempted to save JPEGs at quality 100 (many do).

The shape of the compression graph and this example should convince you to think carefully about the trade-off of compression and quality, and not just to always set the quality at maximum or go for the smallest file size every time.

When to use very high quality JPEG settings

The shape of the compression curve actually changes according to the image being compressed. This happens because the opportunities available to JPEG to save space depend on the actual content of the image: how much variability in colour, tone and brightness is there to contend with?

There are some images which suffer more noticeably from the lossy compression of JPEG, because the compression system has been less successful in finding compression opportunities, or has done something to the image which is more noticeable than normal.

Here are some examples:

• Scenes with very large areas of colour gradient, for example, a panoramic sunset - these will look banded in discrete stripes
• Scenes with large dark areas - these will look blocky
• Scenes with high contrast details, eg. text, logos and sharp boundaries - these will look fuzzy

In images of this sort, you must either use a higher JPEG compression level, or (especially in the case of text and logos) move to a lossless compression format. You might also like JPEG 2000, which is very good at avoiding these kind of problems.

JPEG Image Quality

Now it's time to look at image quality. Compression is only part of image quality, and I want to explore several other facets of quality which contribute to the final result.

A new definition: "perceived quality"

We need to start with a perfect photograph on a perfect DSLR, with a perfect lens. Imagine that the equipment, the lighting, the focus and even the subject are all "just right", and then you hit the shutter and capture the perfect photograph. This starts our discussion with a datum from which we can measure our losses, and avoids problems discussing images that are not high quality sources to start with due to equipment flaws.

The digital camera transforms the light that has hit the sensor into a digitised image. Modern 35mm digital backs are capturing images at around 18 megapixels - that means you have 18 million points of information in the digitised version of the image. One way to think of it is as a square image area of 4200 pixels in both horizontal and vertical space. (4200 x 4200 = about 18 million)

It's important to realise just exactly what an enormous amount of information this is. You'd need two 30" LCD displays operating at 72 dpi to get the image width in view without zooming out. Alternatively, at 300 dpi you could print an image over 14 inches wide from this source. (Confused? Don't forget to read the blog article about image resolution!).

Now let's save the file as a JPEG.

For the sake of argument, we save it at 95% quality. This is the first point in the scenario of the perfect image that we've made a compromise: we've thrown away some information, or diluted the perfection of the photo, by using JPEG.

The chances are we can't see any difference at all. There are two reasons:

1. JPEG compression is very effective at retaining almost all important details in a photograph when used at 95% compression;
2. The original file is so big that the defects and artifacts JPEG has introduced are small in relation to the image area.

There are defects and changes in the JPEG, however, our perception of quality is unchanged.

You could continue to compress the image more and more until you can notice the difference. At that point you have an image that is perceptibly lower quality than the original, perfect photo. That point will be different for every person you test, every viewing system you test (eg. screen or print) and every different photo you test because of the nature of JPEG compression.

When we talk about JPEG quality, we need to be careful to say "perceived quality" and we need to recognise that it is hard to measure in an absolute way.

Image area versus JPEG quality

We're discussing a very large image (4200x4200). Now let's say my intention is to store this image in a way that is suitable for printing at 300 dpi, using JPEG to save on disk space.

As I mentioned above, by dividing the pixels into the DPI value 300 we get a print that is 14" x 14" as our theoretical maximum.

If I save the image at very high compression settings, where the compression artifacts are highly pronounced and the original image is badly degraded, I still have 4200 x 4200 pixels. It's very important to make some allowance for this: the information that has been lost cannot be recovered, and we should accept that the maximum print size is no longer 14" x 14". It must be less.

What is the right balance of image size and compression for high quality JPEGs?

JPEG artifacts are much more noticeable at high compression settings. This argument is most powerful when we are talking about over-compression of files that are too big to start with.

In that case, the best thing to do is resize the source image so that it is smaller, and then save it with a lower compression level. You could actually halve its dimensions - which might sound surprising - and get better perceived quality this way.

Here are some relevant numbers:

• Reducing from 18 megapixels to 9 megapixels is a halving of the image area (a 50% saving in raw data)
• A 9 megapixel image itself is still 3000 x 3000 pixels (ie. 71% of the original width and 71% of the original height).
• The compression level can now be moved away from the settings where JPEG artifacts are very obvious to somewhere where they are more subtle.

I've prepared two example images and magnified them for you to show you the effect.

• Image A's original is exactly 8 megapixels
• Image B's original is exactly 4 megapixels
• Both files have the same size on disk (to within 1%)

(Important: I have zoomed in substantially and cropped a small area to show you the results more clearly)

IMAGE A - 8 megapixel source, very high compression:

IMAGE B - 4 megapixel source, modest compression:

Reference image - No compression:

Remember, image A and image B's sources have the same file size on disk. The lower perceived quality in image A is caused by using too much JPEG compression, even though it's got twice as many pixels as image B.

It is as if we have two competing factors - image compression versus image content - and that the reduction in area approach to saving space wins. This happens because reducing the number of pixels provides a shorter route to obtaining a small file size than the JPEG compression curve can compete with.

Notes: The example uses very heavy compression (image A quality setting = 10 and image B quality setting = 26), and the difference is quite pronounced. In general, the effect could be more subtle. I performed the test with image A = 50 and B = 75 (again giving identical file sizes). Another pairing is A = 80, B = 89 for my test photo.

As the gap closes, the detail in the 8 megapixel file is being more properly preserved. Assuming you are trying to minimise the file size (probably to move the file quickly over the internet), then if you need that detail you should not use excessive compression. If you don't need that detail, reduce the image dimensions first.

Summary: how to define image quality for JPEGs

"Quality" is one of the words which has become embedded in our language when we talk about JPEG images. When we use this word we're normally saying something like this: "high quality means low compression; low quality means high compression".

While that captures the essence of JPEG compression trends, it doesn't connect to our perception of the image quality we obtain quite so neatly.

If you're trying to use JPEG to compress very large files for a substantial saving in disk space, you are probably better advised to reduce the pixel size of the source and use a less aggressive compression setting for that JPEG.

About the Author

The author of this blog is Michael Wells, founder of Third Light Ltd.

Third Light is a Digital Asset Management software and internet services business based in Cambridge, UK.

Our core business is to develop and host IMS (Intelligent Media Server), an cutting-edge system for business users, with an outstanding user interface and rich functionality.

You can get in touch with Michael on LinkedIn.

Thanks for looking at our High Quality information.

You are welcome to see more Slide Trombone information and products.