SRL Archive - Discussions and Debates

<n3ss3s> Oh fuck
<n3ss3s> came up with a system for making ocr super fast
<n3ss3s> as if it was slow but anyways
<n3ss3s> I'm assuming simba does ocr by checking for the letter's array of points sort of like it was a dtm (didnt check sauce, cbf)
<n3ss3s> instead of looping through every single point and checking it
<n3ss3s> we should load these weird arrays that's indices lead you to the right letter
<n3ss3s> like let's say for some font the letter l's point array was (0, 1) (0, 2) (0, 3) (0, 4)
<n3ss3s> there would be a string array with 8 dimensions
<n3ss3s> where
<n3ss3s> arr[0][1][0][2][0][3][0][4] = 'l'
<n3ss3s> —> winning

So to get the letter for a point array, you'd do loadedLetterMatrix[px1][py1][px2]…etc, and if it was a valid array of points, it'd take you to a letter.

It's a good idea, but that's not how Simba does it (as if it were a DTM).
Besides, the slowest part is filtering.

So, Getting the Point count of each Letter/Number position and then basically comparing them two any array's of the same Point count instead of cycling through bitmaps? That would be way quicker..
Omg that beard is so disturbing the dude look's like a Tree sort of or something Ahhh! lol

The OCR uses a simple neural network. The ``hard'' part (or rather, where it goes wrong sometimes) is filtering out colours that are not text. Runescape has been making it harder and harder to do this. There are probably some techniques worth exploring, but I haven't found time to do this.