|
Post by jordan on Nov 26, 2018 5:16:17 GMT
I used sikuli before for ldplayer (a kind of Android emulator).
I can do exact match of images when I capture images with snipping tools from windows, if I set ldplayer to show in original size without scaling.
With Ankulua, I tried the same thing with the following code. It doesn't work all the time.
exists(Pattern("1.png"):similar(1.0)
1 How should I capture the image, so I can use exact match in Ankulua 2 What do I need to put in the script to make it happen? 3 What capture SDK is used in ankulua? Which may give me some more ideas
Thanks
|
|
|
Post by AnkuLua on Nov 26, 2018 6:15:23 GMT
|
|
|
Post by jordan on Nov 27, 2018 2:27:44 GMT
Thanks for reply.
The problem is when I use 0.99. Some images are missing from the recognition. If I give 0.95, some wrong images are recognized.
In sikuli, when I use 0.99 with the image captured with snipping tool. This won't happen.
I used the snap your script with 10 colors, then I do snapshot() in detect code. check exists(Pattern("x.png"):similar(0.99). But it still have the same issue. It's not 100% correct.
What should I do? It is important to be 100% correct recognition
Thanks
|
|
|
Post by crossa on Mar 1, 2022 3:47:28 GMT
This is also happening to me. I am using SnapAndPlay and taking the exact image, but sometimes it doesn't match exactly, so I need to make similarity around (0.5 to 0.7)
I also used Sikuli and usually work for bot in PC Platform, and I can say there is a different performance in matching image.
For me, I usually make a tiny region to match the image. So it will not mismatch other images, and importantly Ankulua will work faster.
But I wondered if there is any other method because sometimes, even in the small region, I don't want to mismatch the images.
|
|
|
Post by Exilereven on Mar 1, 2022 12:32:14 GMT
This is also happening to me. I am using SnapAndPlay and taking the exact image, but sometimes it doesn't match exactly, so I need to make similarity around (0.5 to 0.7) I also used Sikuli and usually work for bot in PC Platform, and I can say there is a different performance in matching image. For me, I usually make a tiny region to match the image. So it will not mismatch other images, and importantly Ankulua will work faster. But I wondered if there is any other method because sometimes, even in the small region, I don't want to mismatch the images. If we could see some example images that your having trouble matching? most the time if they are not the same color you can just do a pixel color grab from the center of the image found, to make sure its the "right"image. there are also Masking techniques. are you using black and white images for the match? try using snapshotColor() and see if it identifies them as seperate images better. you have to kind get tricky sometimes when 2 images are very similar, try some differnt approaches.
|
|
|
Post by crossa on Mar 1, 2022 14:25:52 GMT
This is also happening to me. I am using SnapAndPlay and taking the exact image, but sometimes it doesn't match exactly, so I need to make similarity around (0.5 to 0.7) I also used Sikuli and usually work for bot in PC Platform, and I can say there is a different performance in matching image. For me, I usually make a tiny region to match the image. So it will not mismatch other images, and importantly Ankulua will work faster. But I wondered if there is any other method because sometimes, even in the small region, I don't want to mismatch the images. If we could see some example images that your having trouble matching? most the time if they are not the same color you can just do a pixel color grab from the center of the image found, to make sure its the "right"image. there are also Masking techniques. are you using black and white images for the match? try using snapshotColor() and see if it identifies them as seperate images better. you have to kind get tricky sometimes when 2 images are very similar, try some differnt approaches. I have already posted a thread for the case. ankulua.boards.net/thread/2012/matching-nearly-same-image. In my case, it is not a picture but a text with a number. I am thinking about the OCR thing. But I am not sure. And yeah, I also try some different approaches but still haven't found it.
|
|