Wizzup?
2012-02-01 17:42:42
This is on my to-do list for the Simba Uptext/OCR at the moment:
This thread is for additional suggestions on increasing the accuracy of our filter.
For information on how it currently works, read: http://docs.villavu.com/simba/mmlref/ocr.html#uptext
This thread is also for uptext that isn't currently properly detected by Simba. You can post those here, but they must be made in a specific manner so we can effectively use them. Mainly: The (0, 0) of the screenshot should be the same as the (0, 0) of the RS window / the format must be lossless. (not jpg)
You can use something like this:
You can them remove any references that may lead to your character from the screenshot, or further edit it with GIMP/Photoshop. And by edit I mean, remove most of the pictures contents - however, keep the 0, 0 of the picture the same… if you get what I mean. That's nice for automatic testing.
So let's build a set of well-created screenshots that we can use to improve on the OCR. Once a screenshot if confirmed to work on a stable Simba, I or some other mod will mark it ``fixed'' (by editing the post) - or something along the lines of that.
E: Example of how we could set our own filter: (It's not a very clear example, but this filter is really all there's to it - for the current colour filter in Simba 983 anyway)
Filter per ``base'' colour. If there are too few or too many of them in a TPA, remove them. (Filter additional noise)Filter different colours next to each other(Shadow filter does this <3)- Filter if there simply aren't enough colours of the colour as a whole. (Same after SplitTPA/TPAToATPA)
Allow the script to set / customise the initial colour filter. (We can use this to update the colour filter without having to update simba too)(Mostly done)Create a new and better OCR tool. (Easier for everyone to debug the ocr / testing filters)(Work started, going to add more features later)
This thread is for additional suggestions on increasing the accuracy of our filter.
For information on how it currently works, read: http://docs.villavu.com/simba/mmlref/ocr.html#uptext
This thread is also for uptext that isn't currently properly detected by Simba. You can post those here, but they must be made in a specific manner so we can effectively use them. Mainly: The (0, 0) of the screenshot should be the same as the (0, 0) of the RS window / the format must be lossless. (not jpg)
You can use something like this:
SaveScreenshot('a-name.bmp'); // If you have the current unreleased Simba 0.99 you can use .png too
You can them remove any references that may lead to your character from the screenshot, or further edit it with GIMP/Photoshop. And by edit I mean, remove most of the pictures contents - however, keep the 0, 0 of the picture the same… if you get what I mean. That's nice for automatic testing.
So let's build a set of well-created screenshots that we can use to improve on the OCR. Once a screenshot if confirmed to work on a stable Simba, I or some other mod will mark it ``fixed'' (by editing the post) - or something along the lines of that.
E: Example of how we could set our own filter: (It's not a very clear example, but this filter is really all there's to it - for the current colour filter in Simba 983 anyway)
// load0(r_low, r_high, g_low, g_high, b_low, b_high, set_color_to);
// OF_HN = -1, OF_LN = 256 (basically it means, disregard this)
fd[0] := load0(65, OF_HN, OF_LN, 190, OF_LN, 190, ocr_Blue); // blue
fd[1] := load0(65, OF_HN, OF_LN, 190, 65, OF_HN, ocr_Green); // green
fd[2] := load0(OF_LN, 190, 220, 100, 127, 40, ocr_ItemC); // itemC
fd[3] := load0(OF_LN, 190, OF_LN, 190, 65, OF_HN, ocr_Yellow); // yellow
fd[4] := load0(OF_LN, 190, 65, OF_HN, 65, OF_HN, ocr_Red); // red
fd[5] := load0(OF_LN, 190, OF_LN, 65, 65, OF_HN, ocr_Red); // red 2
fd[6] := load0(190 + 10, 130, OF_LN, 65 - 10, 20, OF_HN, ocr_Green); // green 2
fd[7] := load0(190, 140, 210, 150, 200, 160, ocr_ItemC); // item2, temp item_c
fd[8] := load0(65, OF_HN, 65, OF_HN, 65, OF_HN, ocr_Purple); // shadow


Daemonheim.. as show in the picture above.. Results below:
Zanaris as shown in the picture above: