[Imaging] An objective review of the JPEGmini webtool

There is a webtool that is called “JPEGmini”. You can upload images (JPEGs) to it, and it will reduce them in size. They have a very nice comparison tool, and they state you do not get any noticable loss, and they let you download sample images.
They had a banned with reactions of users saying things like “My 80Gb photo collection has reduced to 20Gb”. That got me a bit scared.
Obviously this tool is super nice for people who don’t know what JPEGs are, and do not have the tools to recompress them (photoshop?) but i doubted it should be used for anything else. So anyway, let’s put it to the test.

Their sample images

You can download their sample images. On their website you cannot see the difference between their 10Mb file and their 1Mb file. Hardly surprising since it’s a 3000×2000 file they display in a 600px wide frame. That way you can’t see the detail that you are losing. On very close inspection in photoshop on 100%, i couldn’t see the difference either, but the difference was just some noise. The catch is that their sample files are crappy quality files that are ideal to be compressed, leading to an insane extra compression of 10Mb to 1Mb. A nice feat, but the compression is lossy (due to the nature of a JPEG). So what do we lose?

Front page

Tests

I have tested a Golden Thread Device Level Target. This target is used for measuring the performance of your camera on a variety of image quality aspects. This target has a few features we will test on.
Let’s test the standard deviation (noise), color encoding, oversharpening and sampling efficiency (or sharpness).

Our sample image test target

Test Method

To test the images, we use the free on-line image quality assessment tool called Deltae (http://delt.ae/). You can just drop your images on the website and it will assess them for you, and the images have to contain a test target.

Retrieving the JPEGmini from their website

I just uploaded the image and got the following file. On first sight, i couldnt see real difference, also, it nicely kept the EXIF info intact.

The image JPEGmini'd.

Uploading the file to the Deltae website.

Uploading is easy, just drop the two files on (the original and the JPEGmini version).

Screenshot from the output of the Deltae website

Results!

The Deltae quality analys tool gave me the following results that I’ve put in a table:

File File size (Mb) Noise (stdev) Color encoding (deltaE2000) Sampling Efficiency (%) Oversharpening (%)
Original 2.7 1.1 2.0 88% 0%
JPEGmini 1.8 1.0 2.0 78% 0%

Conclusion

The color accuracy is very good. it doesn’t significantly change the overall color at all, which can be expected.
The actual noise actually decreases, probably due to a small loss in detail (yes, noise is also detail).
The filesize shrunk 1.4x for this image, but we have seen this result varies per image type.
There seems to extra oversharpening applied to the image, this shouldn’t be there, and wasn’t there. Good!
There seems to be a loss of 10% sharpeness in this image. This is being measured through the five slanted edges. I repeated this for multiple targets with the same results.

This last fact is very worrying. Although your files might ‘look’ the same, do NOT throw away your master files because you are in fact losing information, which you might not see, which is probably because you are not looking at it at 100% anyway. So NEVER use this as a replacement for your files. This service is great and do use it, but NEVER throw away your master files. You might as well just shrink the size of the image.

All in all, a nice website, great that it’s free, but never ever throw away your master files.

This worries me. If the compressy is 'lossy'; you are just throwing away perfectly fine data!

The viewer you get after uploading an image on JPEGmini.

Tim Zaman

MSc Biorobotics. Specialization in computer vision and deep learning. Works at NVIDIA.

3 Responses

  1. Felix says:

    There’s no use for this tool these days. Its compression ratio is too high for photo collections and too low for web publishing. For optimizing web graphics better use http://compressjpg.com

  2. Stb says:

    Well it’s bound to be lossy at 5x the compression ratio. It’s just like mp3 for audio. You might not see/hear the difference but the things you don’t hear / see get thrown away. Is that a loss yes, does it matter, probably not.

    • Tim says:

      It is not compression, rather it’s throwing away information. They advertise this tool for use of compressing your pictures. Now these pictures are unique master files. MP3’s are plain derivatives, but there exist many lossless masterfiles. What JPEGmini promotes is throwing away data, You might as well just buy a cheaper camera. It might “look” the same, but the data becomes useless for any postprocessing or ‘remastering’ to stay in your analog.