Legacy ImageMagick Discussions Archive

Hi!

We are about to switch from ImageMagick 6.9 to 7.0 in our application and so I have done some testing and found a problem in the clipping operation. Usually we take a TIF image with a path and convert it into a PNG with the background removed using the following command:

convert image.tif -alpha transparent -clip -alpha opaque -strip image.png

And here is the problem:
With IM 6 the image was clipped "closer". The edge between the image and the transparency was barely visible. With IM 7 this edge is much more noticeable. For example when I set the PNG image on a black background there are white pixels all around the image.

Here are two examples:

http://www.awg88.de/imagemagick/2016-05-12/image_v6.png
http://www.awg88.de/imagemagick/2016-05-12/image_v7.png

The first one is made with IM ImageMagick-6.9.3-Q16 and the second one with ImageMagick-7.0.1-Q16. You can see the difference when you switch between the two images. There are much more white pixels around the image in the second image.

Here are some questions:
- any suggestions on how to get the same/better results as in IM 6?
- is it possible to adjust the distance between the clip-path and the image?
- is it possible to create the alpha-channel with antialiasing around the edges?

The original TIF image can be downloaded here:

http://www.awg88.de/imagemagick/2016-05-12/image.tif

Any help would be appreciated!

JD

I did some more research on this and got some more information. After reading some other posts I think that the resulting PNG has only 1-bit transparency although 8-bit transparency should be possible (see https://www.imagemagick.org/discourse-s ... hp?t=24176).

But first off all some images to clarify the problem I have got

This is the image I took for the following examples:
http://www.awg88.de/imagemagick/2016-05-26/43924200.tif

When I use Photoshop to remove the background I get smooth, semi transparent edges around the image:



Putting the PNG on a black background gives the intended result:



Then I use ImageMagick to create the PNG with the following command:
The resulting PNG looks like this:



Putting tis PNG on a black background gives this not -so-nice result:



The edges between the image an the background are not smooth.

I found a solution for the problem in this post:

https://www.imagemagick.org/discourse-s ... hp?t=24176

In a first step the clip-path is extracted into a separate SVG-file:
In a second step this SVG is used to remove the background from the image:
The resulting PNG has an 8-bit alpha channel and putting it on a background gives nice smooth edges between the image and the background. Mission accomplished!

The question is: Is there an easier way to achieve this without extracting the path first?

Are you saying that v6 gives the correct result but v7 doesn't? If so, I suggest you submit a bug report. It should contain an easily reproducible case that clearly shows the problem.

The result in V6 was a little better. Not much, but noticable in direct comparison. The edges were not semi-transparent, but closer to the image. More of the white pixels around the image from the original background were removed. I think thats the way the simple "-alpha transparent -clip -alpha opaque" action works in V6 and V7. The result was slightly better in V6 but I wouldn't say that there is an error in V7.

I found a solution that produces a perfect result, but that solution takes two steps (see above post). And that makes it difficult to integrate in our workflow. So I am searching for a way to combine the two steps into one convert-command.

JD

try this, though you still need to extract the clip path
I assume you can pipe in Windows. It works on a Mac (IM 6.9.4.4 Q16)

I also tried:
And then did a compare
883.802 (0.013486)

The result is small, but could be better. I am not sure why the difference.

I believe the original code (from Anthony's pages on clip-path is disabling the anti-aliased pixels and using a binary clipping path; whereas your modified code is keeping the anti-aliases alpha values from the clip-path . This appears to be what is happening if I add +dither -colors 256 and look at the histogram. Your result has grayscale alpha, but the orginal result is binary alpha.

The solution works perfekt and could easily be integrated in our workflow. So thats the way I will go.

Thanks a lot!

JD