Post by SpiderBait on Mar 12, 2017 10:46:42 GMT
I need to find about 30 round character images that appear in varied positions and sizes in a region that comprises about 40% of the screen, all images have a frame in the background for a total of 3 frame types.
So I thought of 3 approaches:
1 - Due to the size variation, extracting the images by traditional methods did not seem to be very efficient, so I thought of using the original characters' images extracted from the application, resized to the average size of the images on the screen and use a similarity of 40%.
But even so it would result in a search of 30 images in a region corresponding to 40% of the screen.
2 - Then thinking about efficiency I tried to focus on what the images had in common, the background frames, although they vary in size in the same proportion of the images they are only 3, but for that it would be necessary to use the pure frame with transparency in the middle, which I do not know if it is possible.
3 - And to finish I thought of simplifying and forgetting the detection of the characters, and try to compare a clean image of the region without any of the characters and one with the current screen and click on any difference, which I also do not know if it is possible.
So the question remains whether any of these approaches is feasible, or is there a more efficient approach?
Samples
*That is the link to extracted images, sample screens and the Apk, the bat file is to compose the original images in the frame cut the center and reduce by 50%, it uses ImageMagick.
So I thought of 3 approaches:
1 - Due to the size variation, extracting the images by traditional methods did not seem to be very efficient, so I thought of using the original characters' images extracted from the application, resized to the average size of the images on the screen and use a similarity of 40%.
But even so it would result in a search of 30 images in a region corresponding to 40% of the screen.
2 - Then thinking about efficiency I tried to focus on what the images had in common, the background frames, although they vary in size in the same proportion of the images they are only 3, but for that it would be necessary to use the pure frame with transparency in the middle, which I do not know if it is possible.
3 - And to finish I thought of simplifying and forgetting the detection of the characters, and try to compare a clean image of the region without any of the characters and one with the current screen and click on any difference, which I also do not know if it is possible.
So the question remains whether any of these approaches is feasible, or is there a more efficient approach?
Samples
*That is the link to extracted images, sample screens and the Apk, the bat file is to compose the original images in the frame cut the center and reduce by 50%, it uses ImageMagick.
