Remove lines and noise in the image

[jueqing1015]

I want to use MAGICK.NET to delete the lines and noises in the image
Try Morphology but the effect is not too obvious

[dlemstra]

The links to your images appear to be broken.

[jueqing1015]

I don’t know why the image link I uploaded always fails, I created a compressed package, you download it and take a look
sample-0005 - 副本 (2).zip

[dlemstra]

I think we need the expert @fmw42 to help us with this. If he can tell us how to do it on the command line I can help you translate it to c#.

[jueqing1015]

@dlemstra
Hello, can you help me translate to C#?

[fmw42]

The best way would be to use connected components to remove very small and very large regions.

This leaves only a couple of small black spots:

convert -quiet sample-0005-original.jpg \
-threshold 0 -type bilevel \
-define connected-components:exclude-header=true \
-define connected-components:area-threshold=3-1000 \
-define connected-components:mean-color=true \
-connected-components 8 \
sample-0005_result1.png

The removes those spots, but fills in white holes in some letters such as "e"

convert -quiet sample-0005-original.jpg \
-threshold 0 -type bilevel \
-define connected-components:exclude-header=true \
-define connected-components:area-threshold=4-1000 \
-define connected-components:mean-color=true \
-connected-components 8 \
sample-0005_result2.png

[dlemstra]

You currently cannot set a range for the area threshold but you will be able to do this when version 7.20.0.0 has been released:

using (var image = new MagickImage(@"sample-0005 - 副本 (2).jpg"))
{
    image.Threshold(new Percentage(0)); // -threshold 0
    image.ColorType = ColorType.Bilevel; // -type bilevel

    var settings = new ConnectedComponentsSettings()
    {
        AreaThreshold = new Threshold(3.05, 1000), // -define connected-components:area-threshold=3.05-1000
        MeanColor = true, // -define connected-components:mean-color=true
        Connectivity = 8, // -connected-components 8
    };
    image.ConnectedComponents(settings); // -connected-components 8

    image.Write("sample-0005-result.jpg");
}

I used 3.05 instead of 4 and got a slightly better result for this image.

[jueqing1015]

I use Magick.NET-Q8-AnyCPU
The strange thing is that some images do not report errors and some report errors.

[fmw42]

Q8 will only allow 255 individual regions in your thresholded image. So you may have to use morphology to combine regions before using connected components. Q16 would allow 65535 regions. So you may want to consider upgrading to Q16 ImageMagick install.

[jueqing1015]

@dlemstra
If upgrading to the Q16 version, will it cause the entire process to be slow, because I see the difference between the two. Q16 requires a larger memory space

[fmw42]

Yes, it will be a bit slower due to the larger memory usage.

[jueqing1015]

It may not be able to upgrade, because the overall speed of Q16 will slow down the entire process. This will affect the efficiency a bit. Thank you!

[fmw42]

Text images like the one above have lots of characters and thus lots of isolated regions. If there are more than 255 characters (including dots on i), then connected components will limit out for Q8. That is the trade you will have to assess vs using Q16. Your processing will likely also slow down if you are using HDRI compile.