I'm wondering about what types of metrics I can use for comparing different (lossy) image compression methods (i.e., things other than compression ratio). For example, comparing JPEG, JPEG 2000, and JPEG XR on a set of different images.
Ideally I'd like to do this in a python notebook, but I'm open to any suggestions.
Thank you!
I think this boils down to comparing the lossy images to their original, i.e. measuring the loss of image quality.
Measure the difference between the lossy image and its origin, do the same for other lossy images and compare the results.
How this can be done has been asked before and I have just added an answer mentioning some approaches.
For example in audio codecs like Opus, MDCT is used with 50% percent overlap to avoid ringing artifacts. Why a similar approach is not used in image codecs. e.g., JPEG uses non-overlapping 8x8 blocks ?
Later lossy image codecs like JPEG2000 do use overlapped transforms, but these techniques just weren't around when JPEG was being defined. The wavelet transform that JPEG2000 is based on hadn't been invented yet, and time-domain anti-aliasing techniques like MDCT were extremely new.
For the MDCT in particular, as far as I know it is not used for image compression at all, even Today. I would guess that's because its basis vectors are asymmetric, which makes it intuitively difficult to choose for imaging applications.
I'm working with lots of camera's which capture in BG bayer pattern natively.
Now, every time I record some data, I save it to the disk in the raw bayer pattern, in an avi container. The problem is, that this really adds up after a while. After one year of research, I have close to 4TB of data...
So I'm looking for a lossless codec to compress this data. I know I could use libx264 (with --qp 0), or huffYUV, dirac or jpeg2000, but they all assume you have RGB or YUV data. It's easy enough to convert the bayered data to RGB, and then compress it, but it kind of defeats the purpose of compression if you first triple the data. This would also mean that the demoasicing artefacts introduced by debayering would also be in my source data, which is also not too great. It would be nice to have a codec that can work on the bayered data directly.
Even more nice would be that the solution would involve a codec that is already supported by gstreamer (or ffmpeg), since that's what I am already using.
A rather late suggestion, maybe useful for others..
It helps to deinterleave the Bayer pattern into four quadrants and then treat that image as grayscale. The sub-images (e.g. all red pixels in top left) have half the spatial resolution, but their pixels are more highly correlated. This leads to lower residuals from predictors using nearby pixels and therefore to better compression ratios.
I've seen this reach 2-3x lossless compression on 12-bit raw camera data.
If a commercial solution is ok, check out Cineform. I've used their sdk for a custom video compressor and it works great plus they have some great tools for processing the raw video.
Or if you prefer the open source route check out Elphel JP4.
All I know about Bayer Patterns I learned from Wikipedia, but isn't conversion to RGB more of a deinterlacing than a tripling? Doesn't the resolution for red and blue go down by a factor of 4 and green by a factor of 2? If so, a lossless image compression scheme like lossless jpeg might be just the thing.
I have some (millions) of 16-bit losslessly compressed TIFFs (about 2MB each) and after exhausting TB of disk space I think it's time I archive the older TIFFs as 8-bit JPEGs. Each individual image is a grayscale image, though there may be as many as 5 such images representing the same imaging area at different wavelengths. Now I want to preserve as much information as possible in this process, including the ability to restore the images to their approximate original values. I know there are ways to get further savings through spatial correlations across multiple channels, but the number of channels can vary, and it would be nice to be able to load channels independently.
The images themselves suggest some possible strategies to use since close to ~60% of the area in each image is dark 'background'. So one way to preserve more of the useful image range is just to threshold away anything below this 'background' before scaling and reducing the bit depth. This strategy is, of course, pretty subjective, and I'm looking for any other suggestions for strategies that are demonstrably superior and/or more general. Maybe something like trying to preserve the most image entropy?
Thanks.
Your 2MB TIFFs are already losslessly compressed, so you would be hard-pressed to find a method that allows you to "restore the images" to their original value ranges without some loss of intensity detail.
So here are some questions to narrow down your problem a bit:
What are the image dimensions and number of channels? It's a bit difficult to guess from the filesize and bit depth alone, because as you've mentioned you're using lossless compression. A sample image would be good.
What sort of images are they? E.g. are they B/W blueprints, X-ray/MRI images, color photographs. You mention that around 60% of the images is "background" -- could you tell us more about the image content?
What are they used for? Is it just for a human viewer, or are they training images for some computer algorithm?
What kind of coding efficiency are you expecting? E.g. for the current 2MB filesize, how small do you want your compressed files to be?
Based on that information, people may be able to suggest something. For example, if your images are just color photographs that people will look at, 4:2:0 chroma subsampling will give you a 50% reduction in space without any visually detectable quality loss. You may even be able to keep your 16-bit image depth, if the reduction is sufficient.
Finally, note that you've compared two fundamentally different things in your question:
"top ~40% of the pixels" -- here it sounds like you're talking about contiguous parts of the intensity spectrum (e.g. intensities from 0.6 to 1.0) -- essentially the probability density function of the image.
"close to ~60% of the area in each image" -- here you're talking about the distribution of pixels in the spatial domain.
In general, these two things are unrelated and comparing them is meaningless. There may be an exception for specific image content -- please put up a representative image to make it obvious what you're dealing with.
If you edit your question, I'll have a look and reply if I think of something.
This is really a two part question, since I don't fully understand how these things work just yet:
My situation: I'm writing a web app which lets the user upload an image. My app then resizes to something displayable (eg: 640x480-ish) and saves the file for use later.
My questions:
Given an arbitrary JPEG file, is it possible to tell what the quality level is, so that I can use that same quality when saving the resized image?
Does this even matter?? Should I be saving all the images at a decent level (eg: 75-80), regardless of the original quality?
I'm not so sure about this because, as I figure it: (let's take an extreme example), if someone had a 5 megapixel image saved at quality 0, it would be blocky as anything. Reducing the image size to 640x480, the blockiness would be smoothed out and barely less noticeable... until I saved it with quality 0 again...
On the other end of the spectrum, if there was an image which was 800x600 with q=0, resizing to 640x480 isn't going to change the fact that it looks like utter crap, so saving with q=80 would be redundant.
Am I even close?
I'm using GD2 library on PHP if that is of any use
You can view compress level using the identify tool in ImageMagick. Download and installation instructions can be found at the official website.
After you install it, run the following command from the command line:
identify -format '%Q' yourimage.jpg
This will return a value from 0 (low quality, small filesize) to 100 (high quality, large filesize).
Information source
JPEG is a lossy format. Every time you save a JPEG same image, regardless of quality level, you will reduce the actual image quality. Therefore even if you did obtain a quality level from the file, you could not maintain that same quality when you save a JPEG again (even at quality=100).
You should save your JPEG at as high a quality as you can afford in terms of file size. Or use a loss-less format such as PNG.
Low quality JPEG files do not simply become more blocky. Instead colour depth is reduced and the detail of sections of the image are removed. You can't rely on lower quality images being blocky and looking ok at smaller sizes.
According to the JFIF spec. the quality number (0-100) is not stored in the image header, although the horizontal and vertical pixel density is stored.
For future visitors, checking the quality of a given jpeg, you could just use imagemagick tooling:
$> identify -format '%Q' filename.jpg
92%
Jpeg compression algorithm has some parameters which influence on the quality of the result image.
One of such parameters are quantization tables which defines how many bits will be used on each coefficient. Different programs use different quatization tables.
Some programs allow user to set quality level 0-100. But there is no common defenition of this number. The image made with Photoshop with 60% quality takes 46 KB, while the image made with GIMP takes only 26 KB.
Quantization tables are also different.
There are other parameters such subsampling, dct method and etc.
So you can't describe all of them by single quality level number and you can't compare quality of jpeg images by single number. But you can create such number like photoshop or gimp which will describe compromiss between size on quality.
More information:
http://patrakov.blogspot.com/2008/12/jpeg-quality-is-meaningless-number.html
Common practice is that you resize the image to appropriate size and apply jpeg after that. In this case huge and middle images will have the same size and quality.
Here is a formula I've found to work well:
jpg100size (the size it should not exceed in bytes for 98-100% quality) = width*height/1.7
jpgxsize = jpg100size*x (x = percent, e.g. 0.65)
so, you could use these to find out statistically what quality your jpg was last saved at. if you want to get it down to let's say 65% quality and if you want to avoid resampling, you should compare the size initially to make sure it's not already too low, and only then reduce the quality
As there are already two answers using identify, here's one that also outputs the file name (for scanning multiple files at once):
If you wish to have a simple output of filename: quality for use on multiple images, you can use
identify -format '%f: %Q' *
to show the filename + compression of all files within the current directory.
So, there are basically two cases you care about:
If an incoming image has quality set too high, it may take up an inappropriate amount of space. Therefore, you might want, for example, to reduce incoming q=99 to q=85.
If an incoming image has quality set too low, it might be a waste of space to raise it's quality. Except that an image that's had a large amount of data discarded won't magically take up more space when the quality is raised -- blocky images will compress very nicely even at high quality settings. So, in my opinion it's perfectly OK to raise incoming q=1 to q=85.
From this I would think simply forcing a decent quality setting is a perfectly acceptable thing to do.
Every new save of the file will further decrease overall quality, by using higher quality values you will preserve more of image. Regardless of what original image quality was.
If you resave a JPEG using the same software that created it originally, using the same settings, you'll find that the damage is minimized - the algorithm will tend to throw out the same information it threw out the first time. I don't think there's any way to know what level was selected just by looking at the file; even if you could, different software almost guarantees different parameters and rounding, making a match almost impossible.
This may be a silly question, but why would you be concerned about micromanaging the quality of the document? I believe if you use ImageMagick to do the conversion, it will manage the quality of the JPEG for you for best effect. http://www.php.net/manual/en/intro.imagick.php
Here are some ways to achieve your (1) and get it right.
There are ways to do this by fitting to the quantization tables. Sherloq - for example - does this:
https://github.com/GuidoBartoli/sherloq
The relevant (python) code is at https://github.com/GuidoBartoli/sherloq/blob/master/gui/quality.py
There is another algorithm written up in https://arxiv.org/abs/1802.00992 - you might consider contacting the author for any code etc.
You can also simulate file_size(image_dimensions,quality_level) and then invert that function/lookup table to get quality_level(image_dimensions,file_size). Hey presto!
Finally, you can adopt a brute-force https://en.wikipedia.org/wiki/Error_level_analysis approach by calculating the difference between the original image and recompressed versions each saved at a different quality level. The quality level of the original is roughly the one for which the difference is minimized. Seems to work reasonably well (but is linear in the for-loop..).
Most often the quality factor used seems to be 75 or 95 which might help you to get to the result faster. Probably no-one would save a JPEG at 100. Probably no-one would usefully save it at < 60 either.
I can add other links for this as they become available - please put them in the comments.
If you trust Irfanview estimation of JPEG compression level you can extract that information from the info text file created by the following Windows line command (your path to i_view32.exe might be different):
"C:\Program Files (x86)\IrfanView\i_view32.exe" <image-file> /info=txtfile
Jpg compression level is recorded in the IPTC data of an image.
Use exiftool (it's free) to get the exif data of an image then do a search on the returned string for "Photoshop Quality". Or at least put the data returned into a text document and check to see what's recorded. It may vary depending on the software used to save the image.
"Writer Name : Adobe Photoshop
Reader Name : Adobe Photoshop CS6
Photoshop Quality : 7"