moparisthebest
/
Simba
mirror of https://github.com/moparisthebest/Simba
1
0
Fork 0
Code Issues Releases Wiki Activity

Clean up ocr.pas a bit.

This commit is contained in:
Merlijn Wajer 2010-09-25 00:37:49 +02:00
parent d54e6ade93
commit 6a0a7f3ebd
1 changed files with 96 additions and 46 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

142
Units/MMLCore/ocr.pas
View File

@ -52,7 +52,7 @@ type
function GetUpTextAtEx(atX, atY: integer; shadow: boolean): string;
function GetUpTextAt(atX, atY: integer; shadow: boolean): string;
procedure FilterUpTextByColour(bmp: TMufasaBitmap; w,h: integer);
procedure FilterUpTextByColour(bmp: TMufasaBitmap);
procedure FilterUpTextByCharacteristics(bmp: TMufasaBitmap; w,h: integer);
procedure FilterShadowBitmap(bmp: TMufasaBitmap);
procedure FilterCharsBitmap(bmp: TMufasaBitmap);
@ -197,84 +197,124 @@ end;
Non optimised. We can make it use direct data instead of fastgetpixel and
fastsetpixel, but speed isn't really an issue. The entire algorithm is still
fast enough.
There are some issues with this method; since later on the algorithm
(if I recall correctly) throws aways some pixels if a character isn't just
one colour)
We will match shadow as well; we need it later on.
}
procedure TMOCR.FilterUpTextByColour(bmp: TMufasaBitmap; w,h: integer);
procedure TMOCR.FilterUpTextByColour(bmp: TMufasaBitmap);
var
x, y,r, g, b: Integer;
begin
// We're going to filter the bitmap solely on colours first.
// If we found one, we set it to it's `normal' colour.
{
We're going to filter the bitmap solely on colours first.
If we found one, we set it to it's `normal' colour.
Note that these values aren't randomly chosen, but I didn't spent too
much time on finding the exact values either.
We will iterate over each pixel in the bitmap, and if it matches any of the
``rules'' for the colour; we will set it to a constant colour which
represents this colour (and corresponding rule). Usually the ``base''
colour.
If it doesn't match any rule, we can safely make the pixel black.
To my knowledge this algorithm doesn't remove any valid points. It does
not remove *all* invalid points either; but that is simply not possible
based purely on the colour. (If someone has a good idea, let me know)
}
for y := 0 to bmp.Height - 1 do
for x := 0 to bmp.Width - 1 do
begin
colortorgb(bmp.fastgetpixel(x,y),r,g,b);
// the abs(g-b) < 15 seems to help heaps when taking out crap points
{
abs(g-b) < 15 seems to help heaps when taking out invalid (white) points
obviously if a colour is ``white'' R, G B should all be approx the
same value. That is what the last part of the if statement checks.
}
if (r > ocr_Limit_High) and (g > ocr_Limit_High) and (b > ocr_Limit_High)
// 50 or 55. 55 seems to be better.
and (abs(r-g) + abs(r-b) + abs(g-b) < 55) then
// TODO: make 55 a var, and make it so that it can be set
begin
bmp.fastsetpixel(x,y,ocr_White);
continue;
end;
{ Quite straightforward and works quite well, afaik }
if (r < ocr_Limit_Low) and (g > ocr_Limit_High) and (b > ocr_Limit_High) then
begin
bmp.fastsetpixel(x,y,ocr_Blue);
continue;
end;
{ Ditto }
if (r < ocr_Limit_Low) and (g > ocr_Limit_High) and (b < ocr_Limit_Low) then
begin
bmp.fastsetpixel(x,y,ocr_Green);
continue;
end;
// false results with fire
{ False results with fire }
if(r > ocr_Limit_High) and (g > 100) and (g < ocr_Limit_High) and (b > 40) and (b < 127) then
begin
bmp.fastsetpixel(x,y,ocr_ItemC);
continue;
end;
{ Works fine afaik }
if(r > ocr_Limit_High) and (g > ocr_Limit_High) and (b < ocr_Limit_Low) then
begin
bmp.fastsetpixel(x,y,ocr_Yellow);
continue;
end;
// better use g < 40 than ocr_Limit_Low imo
// better use g < 40 than ocr_Limit_Low imo (TODO)
if (r > ocr_Limit_High) and (g < ocr_Limit_Low) and (b < ocr_Limit_Low) then
begin
bmp.fastsetpixel(x,y,ocr_Red);
continue;
end;
{ Red as well }
if (r > ocr_Limit_High) and (g > ocr_Limit_Low) and (b < ocr_Limit_Low) then
begin
bmp.fastsetpixel(x,y,ocr_Red);
continue;
end;
if (r > ocr_Limit_Med) and (r < (ocr_Limit_High + 10)) and (g > ocr_Limit_Low - 10) and
(b < 20) then
begin
bmp.fastsetpixel(x,y,ocr_Green);
continue;
end;
//shadow
begin
bmp.fastsetpixel(x,y,ocr_Green);
continue;
end;
// Match Shadow as well. We will need it later.
if (r < ocr_Limit_Low) and (g < ocr_Limit_Low) and (b < ocr_Limit_Low) then
begin
bmp.FastSetPixel(x,y, ocr_Purple);
continue;
end;
// Black if no match
bmp.fastsetpixel(x,y,0);
end;
// make outline black for shadow characteristics filter
// first and last horiz line = 0
{
Make outline black for shadow characteristics filter
First and last horiz line = 0. This makes using the shadow algo easier.
We don't really throw away information either since we already took the
bitmap a bit larger than required.
}
for x := 0 to bmp.width -1 do
bmp.fastsetpixel(x,0,0);
for x := 0 to bmp.width -1 do
bmp.fastsetpixel(x,bmp.height-1,0);
// same for vertical lines
// Same for vertical lines
for y := 0 to bmp.Height -1 do
bmp.fastsetpixel(0, y, 0);
for y := 0 to bmp.Height -1 do
@ -288,7 +328,6 @@ end;
on characteristics.
For the uptext, a few things apply...
First of all:
*** Remove False Shadow ***
if shadow[x,y] then not shadow[x-1,y-1]
@ -300,7 +339,6 @@ end;
will soon see that this algorithm will be a *lot* more efficient if we
start at the right bottom, instead of the left top. Which means we should
work with x-1 and y-1, rather than x+1,y+1
Yeah.... My comments are vague.
)
*** UpText chars identity 1 and 2 ***
@ -314,11 +352,14 @@ procedure TMOCR.FilterUpTextByCharacteristics(bmp: TMufasaBitmap; w,h: integer);
var
x,y: Integer;
begin
// Filter 2
// This performs a `simple' filter.
// What we are doing here is simple checking that if Colour[x,y] is part
// of the uptext, then so must Colour[x+1,y+1], or Colour[x+1,y+1] is a shadow.
// if it is neither, we can safely remove it.
{ Filter 2
This performs a `simple' filter.
What we are doing here is simple checking that if Colour[x,y] is part
of the uptext, then so must Colour[x+1,y+1], or Colour[x+1,y+1] is a shadow.
if it is neither, we can safely remove it.
XXX: This causes the previously mentioned fuckup.
}
for y := 0 to bmp.Height - 2 do
for x := 0 to bmp.Width - 2 do
begin
@ -326,11 +367,20 @@ begin
continue;
if bmp.fastgetpixel(x,y) = clBlack then
continue;
if (bmp.fastgetpixel(x,y) <> bmp.fastgetpixel(x+1,y+1)) and (bmp.fastgetpixel(x+1,y+1) <> clpurple) then
if (bmp.fastgetpixel(x,y) <> bmp.fastgetpixel(x+1,y+1)) and
(bmp.fastgetpixel(x+1,y+1) <> clpurple) then
bmp.fastsetpixel(x,y,{clAqua}0);
end;
// Remove false shadow
{
Remove false shadow. (A shadow isn't larger than 1 pixel. So if x, y is
shadow, thex x+1, y+1 can't be shadow. If it is, we can safely remove it.
(As well as x+2, y+2, etc).
The tricky part of the algorithm is that it starts at the bottom,
removing shadow point x,y if x-1,y-1 is also shadow.
}
for y := bmp.Height - 1 downto 1 do
for x := bmp.Width - 1 downto 1 do
begin
@ -346,24 +396,24 @@ begin
end;
// Now we do another filter, with uptext chars identity 1 and 2.
for y := bmp.Height - 2 downto 0 do
for x := bmp.Width - 2 downto 0 do
begin
if bmp.fastgetpixel(x,y) = clPurple then
continue;
if bmp.fastgetpixel(x,y) = clBlack then
continue;
for y := bmp.Height - 2 downto 0 do
for x := bmp.Width - 2 downto 0 do
begin
if bmp.fastgetpixel(x,y) = clPurple then
continue;
if bmp.fastgetpixel(x,y) = clBlack then
continue;
// identity 1
if (bmp.fastgetpixel(x,y) = bmp.fastgetpixel(x+1,y+1) ) then
continue;
// identity 1
if (bmp.fastgetpixel(x,y) = bmp.fastgetpixel(x+1,y+1) ) then
continue;
// identity 2
if bmp.fastgetpixel(x+1,y+1) <> clPurple then
begin
bmp.fastsetpixel(x,y,clOlive);
continue;
end;
// identity 2
if bmp.fastgetpixel(x+1,y+1) <> clPurple then
begin
bmp.fastsetpixel(x,y,clOlive);
continue;
end;
// If we make it to here, it means the pixel is part of the uptext.
end;
@ -505,7 +555,7 @@ begin
{$ENDIF}
// Filter 1
FilterUpTextByColour(bmp,w,h);
FilterUpTextByColour(bmp);
{$IFDEF OCRSAVEBITMAP}
bmp.SaveToFile(OCRDebugPath + 'ocrcol.bmp');
{$ENDIF}
@ -542,8 +592,8 @@ begin
// TODO:
// We should make a different TPA
// for each colour, rather than put them all in one. Noise can be a of a
// differnet colour.
// for each colour, rather than put them all in one. Noise can be of a
// different colour.
setlength(chars, charsbmp.height * charsbmp.width);
charscount:=0;
for y := 0 to charsbmp.height - 1 do