I see it as available (using gandi.net), but the price is through the roof for this top level domain name. I hope Nigerians can have a better deal, because more than 700 euros a year is crazy even for western pockets.
Cochonours
kip
ex_cldr Core Team
Always open to ideas so if you have a need, open an issue! Or discuss here, happy either way.
sodapopcan
Essentially I need to do @kwando’s example, so an image, resize it, maybe filter it, then compose it together with a mask. I suppose that could be achieved through multiple calls to Imaginary (sorry, thinking in real time here and trying to disaccociate my thinking from what I’ve already been building).
I also need to produce some vector images (for print files) but that is obviously out of scope of Image.
kip
ex_cldr Core Team
That should all be possible in a single web request since transforms will be composed and the URL format allows specifying multiple transforms.
kip
ex_cldr Core Team
I’ve published Image version 0.33.0 with the following breaking changes:
Breaking Changes
-
Image.dhash/1now returns a 64-bit hash rather than a 512-bit hash. The original algorithm is intended to return a 64-bit hash so this is both a breaking change and bug fix. -
Fix generation of images from text when
autofit: false(the default). In these cases the text image is generated from SVG. The original implementation operated on the basis that the image would be generated with a height and width necessary to contain the text of a given font size. The result is not consistent and is influenced by whether the text has ascenders or desenders. The new implementation creates an SVG container of a size large enough to contain the text at the given font-size and then trims away the background to produce the final result. Thanks to @ilesar for the report. Closes #86. As a result of this fix, existing text-generated images may not now be the same as previous releases.
I apologise especially for the breaking changes in text image generation but the previous approach was delivering unpredictable - and usually inconsistent - results.
sodapopcan
eahanson
I tried to use the Image library to do a fuzzy comparison of two images by using Image.hamming_distance but it was too fuzzy for my purposes — a picture of a cat and a picture of a house got a hamming distance of 0 (they were both relatively small, square, grayscale images).
Any suggestions on a way to do a fuzzy comparison that’s a bit less fuzzy? (Maybe something analogous to the ImageMagick command magick compare -metric AE -fuzz 10% cat.png house.png diff.png ?)
kip
ex_cldr Core Team
@eahanson, image comparisons are tricky things and I’m learning a lot. If I’ve understood correctly, the magic compare -metric AE -fuzz 10% means;
-metric AE: means absolute error (ie pixel by pixel comparison)-fuzz 10%means the root mean squared difference between the two images should fall in that range of 10% difference
I use a very similar approach in the test suite so I think implementing this is straight forward. But I’d like to do some validation.
May I ask you to DM me the two images you used (I know they are just examples, but it gives us a baseline) - either attach the images or send links is fine.
I am also going to implement at least PSNR and SSIM sooner rather than later. I’m open to other ideas and suggestions.
kip
ex_cldr Core Team
I’ve published image version 0.34.0. Special shoutout to @sodapopcan who demonstrated amazing patience while I fuddled though making text images work consistently and correctly.
Example of Image.compare/3
The main feature of this release is a way to compare images and return a similarity score and a difference image. This is similar to the same function in ImageMagick.
iex> image_1 = Image.open!("./test/support/images/Kamchatka-2019-8754.jpg")
# Image 2 is the same image, only with a cat overlayed upon it.
iex> cat = Image.open! "./test/support/images/cat.png"
iex> image_2 = Image.compose!(i, cat, x: :middle, y: :center) |> Image.flatten!
# Now we can compare the image with the defaults of
# `metric: :ae, difference_color: :red, saturation: 1.0, brightness: 1.0, difference_boost: 1.5`
iex> {:ok, metric, difference_image} = Image.compare(image_1, image_2)
{:ok, 0.13152029520295203,
%Vix.Vips.Image{ref: #Reference<0.1501890617.2351562776.180408>}}
# Note how the pixels that were different between the two images are masked in red overlaid on the original image. Its possible to adjust the brightness and saturation of the base image and to boost them for the difference pixels. Here's an example:
iex> {:ok, metric, difference_image} = Image.compare image_1, image_2, brightness: 0.5, saturation: 0.2, difference_boost: 2.0
{:ok, 0.13152029520295203,
%Vix.Vips.Image{ref: #Reference<0.1501890617.2351562776.180435>}}
Note how the difference pixels stand out more strongly. You can also see there is a color gradient to the difference pixels. The shading indicates how much different (mathematical difference, not perceptual difference) the pixels are.
11