Character Recognition or OCR Advice

[eskermeko]

Hi

This is my first post here and I would just like to write how wonderful this software is. Its helped me alot.

I have been writing my first OCR. But I am using PixelChecksum and with a gradient background on the pictures I am finding it very hard.

Was wondering can I count the pixels of a certain colour in a specified block of a picture.

I would like to count the white pixels.

Any help welcome.

Edited April 10, 2007 by eskermeko

[PsaltyDS]

Since you didn't get any inteligent answers, I'll give you one of mine instead...

OCR is hard, very hard. Differences in fonts, font size, bold, italics, underlining, color, superscript, subscript, etc., etc. make it an extremely hard thing to code. If you use the Search on this forum for OCR, you'll get a lot of discussion about the dificulties. You could, just, create OCR functions in AutoIT. But they will be very limited, to a particular font and size for example, and likely very slow. Search will find you some examples, but I think you will find they were coded for very limited special circumstances.

For some idea of the level of effort that goes into and OCR project, take a look at Tesseract or the more recent OCRopus projects.

As for your actual AutoIT question, you can walk through a graphic space on the screen with For/Next loops doing PixelGetColor() and count pixels with a given color. You can even do it inside a BMP file of an image fairly straight forwardly. But interpreting what you get will become horribly complicated realy quickly, so think about other ways to get whatever it is done, too.

Edited April 11, 2007 by PsaltyDS

[eskermeko]

Since you didn't get any inteligent answers, I'll give you one of mine instead...

OCR is hard, very hard. Differences in fonts, font size, bold, italics, underlining, color, superscript, subscript, etc., etc. make it an extremely hard thing to code. If you use the Search on this forum for OCR, you'll get a lot of discussion about the dificulties. You could, just, create OCR functions in AutoIT. But they will be very limited, to a particular font and size for example, and likely very slow. Search will find you some examples, but I think you will find they were coded for very limited special circumstances.

For some idea of the level of effort that goes into and OCR project, take a look at Tesseract or the more recent OCRopus projects.

As for your actual AutoIT question, you can walk through a graphic space on the screen with For/Next loops doing PixelGetColor() and count pixels with a given color. You can even do it inside a BMP file of an image fairly straight forwardly. But interpreting what you get will become horribly complicated realy quickly, so think about other ways to get whatever it is done, too.

Thanks I quickly put this together

$xstart = 253
$ystart = 64
$xend = 262
$yend = 74



$x = $xstart
$y = $ystart


$count = 0
while $xend > $x    
        
if PixelSearch( $x, $y, $y+1, $y+1, 0xFF0000 ) Then
    $count = $count + 1
    EndIf


$x = $x + 1
$y = $y +1

Wend

MsgBox(0,"Count=", $count)

Im new at this. Somethings wrong, dunno what.

I keep on getting same number(9) when I count.

Edited April 11, 2007 by eskermeko

[eskermeko]

Oops no wait thats totally wrong ,Think PixelgetColour right. I realized the Pixelsearchcolor should be:

$pixel = PixelSearch( $x, $y, $y+1, $y+1, 0xFFCC00 )

If Not @error Then

$count = $count + 1

EndIf

And Im doing the x y wrong also

Edited April 11, 2007 by eskermeko

[eskermeko]

Well I had to do this:

while $yend > $y

  $z = $y + 1

while $xend > $x
        

$var = PixelGetColor( $x , $z )

If $var = "16763904" then
    $count = $count + 1
    EndIf


$x = $x + 1


    Wend
    $x = $xstart
    $y = $y + 1
    
Wend

[PsaltyDS]

If you will look for the pixels more than once, you want it in a function you call over and over again. Here is an example of search a rectangular area for three different colors:

; Count pixels of a certain color in a 100x40 box starting at 200,300
Dim $xstart = 200, $width = 100
Dim $ystart = 3000, $height = 40

$SearchColor = 0
$Result = _PixelGetColorCount($xstart, $ystart, $width, $height, $SearchColor)
MsgBox(64, "Results", "Black pixels found:  " & $Result)

$SearchColor = 0xFFFFFF
$Result = _PixelGetColorCount($xstart, $ystart, $width, $height, $SearchColor)
MsgBox(64, "Results", "White pixels found:  " & $Result)

$SearchColor = 16763904
$Result = _PixelGetColorCount($xstart, $ystart, $width, $height, $SearchColor)
MsgBox(64, "Results", "Pixels found of color '16763904':  " & $Result)

;-----------------------------
; Function _PixelGetColorCount()
;   Returns the number of pixel of a certain color
;   Call with:  _PixelGetColorCount($x, $y, $w, $h, $color)
;   Returns count of matching pixels on success
;-----------------------------
Func _PixelGetColorCount($x, $y, $w, $h, $color)
    Local $count = 0, $IntX, $IntY
    For $IntX = $x To $x + $w
        For $IntY = $y To $y + $h
            If PixelGetColor($IntX, $IntY) = $color Then $count += 1
        Next
    Next
    Return $count
EndFunc   ;==>_PixelGetColorCount

Edit: From another post, there is a reference to the MODI OCR, which is the Microsoft Office Document Imaging interface. Gives you some more options if you have MS Office 2003 or later installed, but would require learning to use the COM object capabilities of AutoIT to take advantage of.

Edited April 11, 2007 by PsaltyDS

[eskermeko]

THanks I was working on this today

What a mission.

But all for nothing cause some of the letters has the same amount of pixels.

Maybe I can search by shade of colours

With the code I scanned every vertical line and when I found a blank vertical I would reset the counters for a new Letter.

So it doesn't just read one letter.

expandcollapse popup

$xstart = 161
$ystart = 21
$yend   = 31
$count = 0
$height = $yend - $ystart
$width = 24
$colour = "16763904"

$x = $xstart
$y = $ystart



$count2 = 0
$crashProtect = 0

FOR $a = $width to 1 Step -1    

;Init count
$count = 0
;count Vertical Lines 
$count = CountVertical($x,$y,$height,$colour)
;Move on xaxis
$x = $x + 1


;Add Vertical lines together
$count2 = $count2 + $count

;Print & reset counters on new line
IF $count = 0 THEN 
    
    IF $count2 > 1 THEN
ControlSend("Untitled - Notepad", "", "Edit1", " . ") 
ControlSend("Untitled - Notepad", "", "Edit1", $count2)
ControlSend("Untitled - Notepad", "", "Edit1", " . ") 
    ENDIF
    
    $a = $width
        $crashProtect = $crashProtect + 1
         $count2 = 0
Else
    $crashProtect = 0
ENDIF


;Protect against infinite loop
IF $crashProtect > 25 Then
    ExitLoop
EndIf

Next


    


;Functions
Func CountVertical($x,$y,$height,$colour)

    
FOR $i = $height to 1 Step -1   

$var = PixelGetColor( $x , $y )


If PixelGetColor($x, $y) = $colour Then $count += 1


$y = $y + 1

Next

return $count

EndFunc

[eskermeko]

I changed the function to see if I can get shades

But this does not seem to work

;Functions
Func CountVertical($x,$y,$height,$colour)

$z = $y 
FOR $i = $height to 1 Step -1   
$z = $z + 1

$var = PixelSearch($x, $y,$x,$z,$colour, 10) 
If Not @error Then
  $count += 1
EndIf

Next

return $count

EndFunc

Am I doin it right?

Any help appreciated

[eskermeko]

I changed the function to see if I can get shades

But this does not seem to work

;Functions
Func CountVertical($x,$y,$height,$colour)

$z = $y 
FOR $i = $height to 1 Step -1   
$z = $z + 1

$var = PixelSearch($x, $y,$x,$z,$colour, 10) 
If Not @error Then
  $count += 1
EndIf

Next

return $count

EndFunc

Am I doin it right?

Any help appreciated

[eskermeko]

I changed the function to see if I can get shades

But this does not seem to work

;Functions
Func CountVertical($x,$y,$height,$colour)

$z = $y 
FOR $i = $height to 1 Step -1   
$z = $z + 1

$var = PixelSearch($x, $y,$x,$z,$colour, 10) 
If Not @error Then
  $count += 1
EndIf

Next

return $count

EndFunc

Am I doin it right?

Any help appreciated

[eskermeko]

Something went wrong with browser and that s why it printed three posts

[PsaltyDS]

I changed the function to see if I can get shades

But this does not seem to work

Am I doin it right?

Any help appreciated

This works to count one vertical stripe of pixels:

;Functions
Func CountVertical($x, $y, $height, $colour)
    Local $PixY, $count = 0

    For $PixY = $y To $y + $height
        PixelSearch($x, $PixY, $x, $PixY, $colour, 10)
        If Not @error Then $count += 1
    Next

    Return $count
EndFunc   ;==>CountVertical

Something went wrong with browser and that s why it printed three posts

You realize writing your own web browser might be easier than your own OCR?

[eskermeko]

Thanks very much

LOL!

I have written a simple web browser with C# but this is insane man

Edited April 11, 2007 by eskermeko

[eskermeko]

Thanks for the help, I finally got it working.

I was adding the vertical lines all this time instead of just putting them next to each other. That fixed the whole thing.

It was pretty frustrating and hard to do but in the end worth it.

[seangriffin]

You could check out my Tesseract OCR UDF if you are after a free OCR solution:

Tesseract UDF

Cheers,

Sean.

Edited February 16, 2009 by seangriffin

[Marlo]

2 year bump? nice o_0

[dgm5555]

I've just uploaed this native autoit screen OCR (working reasonably well, but still not totally bug-free):-

Also this can search for a pixel pattern (eg a word), but doesn't return the actual text:

May be useful.

David

[BrewManNH]

@dgm5555

Why did you post this here? This thread is 5 years dead, with a 3 year old necroing attempt on it already. Please don't do that again.

[Melba23]

dgm5555,

You have necro-posted twice today - and both times hijacked the threads to publicize your own. Please do not do it a third time.

M23