[HN Gopher] The New Test Images - Image Compression Benchmark
___________________________________________________________________
The New Test Images - Image Compression Benchmark
Author : Zababa
Score : 6 points
Date : 2024-09-02 14:19 UTC (3 days ago)
(HTM) web link (imagecompression.info)
(TXT) w3m dump (imagecompression.info)
| Zababa wrote:
| A bit more context from the page:
|
| > The images historically used for compression research (lena,
| barbra, pepper etc...) have outlived their useful life and its
| about time they become a part of history only. They are too
| small, come from data sources too old and are available in only
| 8-bit precision.
|
| > These high-resolution high-precision images have been carefully
| selected to aid in image compression research and algorithm
| evaluation. These are photographic images chosen to come from a
| wide variety of sources and each one picked to stress different
| aspects of algorithms. Images are available in 8-bit, 16-bit and
| 16-bit linear variations, RGB and gray.
|
| > You are encouraged to use these images for image compression
| research and algorithm evaluation. Suggestions for further
| improvements are always welcome.
|
| I am currently looking at different image formats/png optimizers
| and trying to compare them, so having a set of images to compare
| is nice.
|
| I do feel like they're not exactly representative of what an
| image can be on the web or on my computer. I was thinking of
| adding a few things for my own use: a screenshot, an image macro,
| a page of manga.
| brudgers wrote:
| I am not defending the use of Lena nor advocating for its
| continued use.
|
| A similarly useful replacement should probably contain similar
| types and degrees of detail so that image algorithms can be
| tested. The hair, skin, fabric etc. in Lena readily surface an
| algorithm's compression and convolution artifacts. The Lena
| image was chosen in part for its technical properties.
|
| Unlike the Utah Teapot also used for computer imaging, the Lena
| image is socially and academically inappropriate. Though
| suitable replacements can be found or made, most images are not
| suitable technical replacements. Doing so will probably be non-
| trivial due to the volume of existing technical decisions based
| on the Lena image.
| https://www.tandfonline.com/doi/full/10.1080/09500340.2016.1...
| andrewla wrote:
| I understand why it can be problematic to use humans in test
| images, but this is in conflict with the purpose of test images.
| The purpose, briefly, is to allow human eyes to easily spot
| defects in image processing. While we can use automated methods
| as well especially for formal analysis of fidelity, in the end
| the reason why we are displaying images is for human consumption.
|
| We don't need to to Lena or the other traditional images, but it
| should be possible to find a usable example -- even if it's just
| the grimacing guy from the meme or something.
| refulgentis wrote:
| It's not that it's problematic to use humans, it's that it's
| problematic to use images from 1970s Playboy.
|
| There's a theoretical argument to make that this rando site
| should also have a picture of a human, but it's weak --
| something something evolution over millenia, something
| something parsing faces, therefore we'll notice if its off
| quicker.
|
| It's a weak argument because we don't expect ex. facial
| symmetry to change, and faces are capable of being quite
| asymmetrical.
|
| In practice its _much_ more valuable for the random website to
| have ex. the color test card, deep dark night with streetlights
| to stretch dynamic range, depth of field and details of fine
| spider web above blurry background, etc. It 's fairly easy to
| find a picture of a human when needed, especially one that
| we're familiar with and see small details change :)
| dylan604 wrote:
| Part of the use of having a human was meant to see how well
| skin tones looked after being developed. The big problem was
| that it was only a single skin tone that was ever used.
| Lerc wrote:
| That seems like a poor argument for Lina given how off the
| colour balance is. It is a very pink image.
|
| The merit comes more from the combination of sharp edges,
| gradients, and details that humans are sensitive to (face)
| and details that they are not (feathers).
|
| Spending bits to place an eye at just the right place is more
| important than spending the same bits to place a bit of fluff
| covering the same number of pixels.
|
| Something like the photo in test card F at high resolution
| would be good. (maybe with a fluffier toy)
| hoten wrote:
| More context, here is what the old "Lena" test image is all
| about: https://womenlovetech.com/losing-lena-why-we-need-to-
| remove-... (nsfw, it's a Playboy pic)
| munchler wrote:
| > each one picked to stress different aspects of algorithms
|
| Really? Photographically, these images are kinda crappy - the
| kind of casual snapshot one might take and then delete. The
| cathedral image, for example, is both underexposed and
| overexposed. I guess this isn't too surprising, since it was
| taken on a Nikon D70 from 2004, which had pretty limited dynamic
| range. This hardly seems like a good example to choose for
| testing image compression, since it lacks a lot of useful detail.
| Or maybe this was a deliberate choice? What were the actual
| criteria used to choose these images?
| 1970-01-01 wrote:
| I totally agree. None of them show passionate photography, and
| I hope that's not intentional, as photography is an art. We
| should at minimum benchmark photography that is stimulating for
| a human, otherwise what is the point to any of this?
| entropie wrote:
| is nightshot_iso_1600 and nightshot_iso_100 interchanged? that
| looks not right to me. Iso 100 looks way brighter.
| throwaway7679 wrote:
| (2008)
| sliq wrote:
| This page was last updated in 2015, 9 years ago, and ALOT has
| happened in image compression and on the web in these 9 years.
| And the samples are just so totally weird, they are not
| representative for anything.
___________________________________________________________________
(page generated 2024-09-05 23:01 UTC)