Counting dots in an image (birds, reindeer...)

[Chris Holbrook]

Looking at a picture of a flock of birds on the wing, I wondered how the image could be analysed to count them - black shapes on a blue background. I'm sure that this is used in industry and in science - but what is the algorithm?

The black shapes are of course irregular - but let's simplify it to say that one black shape = 1 reindeer. I mean bird. More sherry, Santa?

[Paul Dixon]

Chris,
if the contrast is good enough then try this code. It was written to count storm systems in a satellite picture but it should be easily modified to count birds (small radius) or swarms (larger radius) if they have enough contrast against the background.
The attached .GIF needs to be converted to a .BMP file and placed in the c:\ directory as storms5.bmp for the demonstration to work as it's written.
Press a key to do each step of the process in turn and display the results.

Paul.

'PBCC5 program
#BREAK ON
#INCLUDE "win32api.inc"

%showscreenupdate=1

FUNCTION PBMAIN() AS LONG
REGISTER pix1 AS LONG
REGISTER y AS LONG

Storm$="c:\storms5.bmp"
radius&=1


'get the size of the bitmap
nFile& = FREEFILE
OPEN Storm$ FOR BINARY AS nFile&
GET #nFile&, 19, nWidth&
GET #nFile&, 23, nHeight&
CLOSE nFile&

'load the bitmap of the storm system
GRAPHIC BITMAP LOAD Storm$,nWidth&,nHeight&,%BLACKONWHITE TO hBmp1???

GRAPHIC ATTACH hBmp1???,0
GRAPHIC GET BITS TO Image1$

xsize& = CVL(image1$,1)
ysize& = CVL(image1$,5)

'show it on screen so I can see the effect of each stage of processing
GRAPHIC WINDOW "Storm ID",1,1,xsize&,ysize& TO hWin1???
GRAPHIC ATTACH hWin1???,0 ,REDRAW

GRAPHIC COPY hBmp1???,0

DIM bmp(xsize&-1,ysize&-1) AS LONG AT STRPTR(Image1$)+8

PRINT xsize&,ysize&
PRINT LEN(image1$)

GRAPHIC REDRAW

CONSOLE SET LOC 800,0
CONSOLE SET FOCUS

PRINT "Original image"
WAITKEY$
PRINT "Removing non black/white pixels and improving contrast"
'remove all non black/white pixels
FOR x& = 0 TO xsize&-1
    FOR y& = 0 TO ysize&-1
        pix1&=bmp(x&,y&)

        r&=pix1& AND 255
        g&=(pix1& AND &h0000ff00) \256
        b&=(pix1& AND &h00ff0000) \ 65536
        brightness& = r&+b&+g&

        IF brightness& > 700 THEN
            GRAPHIC SET PIXEL (x&,y&),%WHITE
            bmp(x&,y&)=%WHITE
        ELSE
            GRAPHIC SET PIXEL (x&,y&),%BLACK
            bmp(x&,y&)=%BLACK

        END IF
    NEXT
#IF %DEF (%showscreenupdate)
    GRAPHIC REDRAW
#ENDIF

NEXT
GRAPHIC REDRAW

WAITKEY$
PRINT "Merging storms within range into systems"

'find those storms within the specified range and merge them into systems
FOR x& = 0 TO xsize&-1
    FOR y& = 0 TO ysize&-1
       ' graphic get pixel (x&,y&) to pix1&
        pix1&=bmp(x&,y&)
        IF pix1& = %WHITE THEN
            GRAPHIC ELLIPSE (x&-radius&,y&-radius&) - (x&+radius&,y&+radius&) ,%YELLOW,%YELLOW,0

        END IF

    NEXT

#IF %DEF (%showscreenupdate)
    GRAPHIC REDRAW
#ENDIF

NEXT
GRAPHIC REDRAW

WAITKEY$
PRINT "Identifying individual systems"
'identify each system from the others
NextColour&=1
DIM systemlocationx(10000) AS LONG
DIM systemlocationy(10000) AS LONG

FOR x& = 0 TO xsize&-1
    FOR y& = 0 TO ysize&-1
        GRAPHIC GET PIXEL (x&,y&) TO pix1&
        IF pix1& = %YELLOW THEN
            GRAPHIC PAINT (x&,y&),NextColour&,%BLACK
            systemlocationx(NextColour&)=x&
            systemlocationy(NextColour&)=y&

            NextColour&=NextColour&+1
        END IF
    NEXT

#IF %DEF (%showscreenupdate)
    GRAPHIC REDRAW
#ENDIF

NEXT
GRAPHIC REDRAW

PRINT NextColour&-1;"Sytems found"
WAITKEY$
PRINT "Calculating size and centre of each system"
SystemLimit&=NextColour&

'find centre of each system
DIM CofGx(SystemLimit&) AS SINGLE
DIM CofGy(SystemLimit&) AS SINGLE
DIM CofGpoints(SystemLimit&) AS SINGLE


FOR x& = 0 TO xsize&-1
    FOR y& = 0 TO ysize&-1
        pix1&=bmp(x&,y&)
        IF pix1& = %WHITE THEN

            GRAPHIC GET PIXEL (x&,y&) TO pix1&
            CofGx(pix1&)=CofGx(pix1&)+x&
            CofGy(pix1&)=CofGy(pix1&)+y&
            CofGpoints(pix1&) = CofGpoints(pix1&)+1
        END IF
    NEXT
NEXT
GRAPHIC REDRAW

PRINT "Removing systems smaller than some limit (for now make it 1000 pixels)"
MaxSize&=0
FOR r&= 1 TO SystemLimit&
    IF CofGpoints(r&)>MaxSize& THEN MaxSize&=CofGpoints(r&)
NEXT
PRINT "Largest System is";MaxSize&;"points"

FOR r&= 1 TO SystemLimit&
    IF CofGpoints(r&)< 1000 THEN
        CofGpoints(r&)=0
        GRAPHIC PAINT (systemlocationx(r&),systemlocationy(r&)), %BLACK,%BLACK
    END IF
NEXT
GRAPHIC REDRAW


WAITKEY$
PRINT "Find edges of remaining storm systems"

'edge detect the systems and draw the yellow line
FOR x& = 1 TO xsize&-2
    FOR y& = 1 TO ysize&-2
        GRAPHIC GET PIXEL (x&,y&) TO pix1&

        IF pix1& = %BLACK THEN
            GRAPHIC GET PIXEL (x&-1,y&-1) TO pix1& :IF pix1&<>%BLACK AND pix1&<>%YELLOW THEN GRAPHIC SET PIXEL (x&,y&),%YELLOW:EXIT IF
            GRAPHIC GET PIXEL (x&  ,y&-1) TO pix1& :IF pix1&<>%BLACK AND pix1&<>%YELLOW THEN GRAPHIC SET PIXEL (x&,y&),%YELLOW:EXIT IF
            GRAPHIC GET PIXEL (x&+1,y&-1) TO pix1& :IF pix1&<>%BLACK AND pix1&<>%YELLOW THEN GRAPHIC SET PIXEL (x&,y&),%YELLOW:EXIT IF
            GRAPHIC GET PIXEL (x&-1,y&  ) TO pix1& :IF pix1&<>%BLACK AND pix1&<>%YELLOW THEN GRAPHIC SET PIXEL (x&,y&),%YELLOW:EXIT IF
            GRAPHIC GET PIXEL (x&+1,y&  ) TO pix1& :IF pix1&<>%BLACK AND pix1&<>%YELLOW THEN GRAPHIC SET PIXEL (x&,y&),%YELLOW:EXIT IF
            GRAPHIC GET PIXEL (x&-1,y&  ) TO pix1& :IF pix1&<>%BLACK AND pix1&<>%YELLOW THEN GRAPHIC SET PIXEL (x&,y&),%YELLOW:EXIT IF
            GRAPHIC GET PIXEL (x&  ,y&+1) TO pix1& :IF pix1&<>%BLACK AND pix1&<>%YELLOW THEN GRAPHIC SET PIXEL (x&,y&),%YELLOW:EXIT IF
            GRAPHIC GET PIXEL (x&+1,y&+1) TO pix1& :IF pix1&<>%BLACK AND pix1&<>%YELLOW THEN GRAPHIC SET PIXEL (x&,y&),%YELLOW:EXIT IF


        END IF
    NEXT

#IF %DEF (%showscreenupdate)
    GRAPHIC REDRAW
#ENDIF

NEXT
GRAPHIC REDRAW
WAITKEY$
PRINT "overlay storm systems with original image"


'GOTO skip    'remove this GOTO if you want the output to look like the original with the clouds in the final picture (as your first example)
             'leave it in if you want the output to just include the system outlines, not the clouds within (as you second example)

'remove all the unwanted colours and overlay with original bitmap
FOR x& = 0 TO xsize&-1
    FOR y& = 0 TO ysize&-1
        GRAPHIC GET PIXEL (x&,y&) TO pix1&
        IF (bmp(x&,y&)=%WHITE) AND (pix1& <> %BLACK) THEN

                GRAPHIC SET PIXEL (x&,y&),%WHITE

        ELSE

            IF pix1& <> %YELLOW THEN
                GRAPHIC SET PIXEL (x&,y&),%BLACK
            END IF
        END IF



    NEXT

#IF %DEF (%showscreenupdate)
    GRAPHIC REDRAW
#ENDIF

NEXT
GRAPHIC REDRAW
WAITKEY$
GOTO skip2

skip:
'Instead of an overlay, just remove the unwanted colours :remove all the unwanted colours and overlay with original bitmap
FOR x& = 0 TO xsize&-1
    FOR y& = 0 TO ysize&-1
        GRAPHIC GET PIXEL (x&,y&) TO pix1&

        IF pix1& <> %YELLOW THEN GRAPHIC SET PIXEL (x&,y&), %BLACK

    NEXT

#IF %DEF (%showscreenupdate)
    GRAPHIC REDRAW
#ENDIF

NEXT
GRAPHIC REDRAW
WAITKEY$

skip2:


PRINT "Show storm centres"

'plot the centre of each system
FOR r& = 1 TO SystemLimit&
    SystemSize&=CofGpoints(r&)
    BlueDotSize&=SQR(CofGpoints(r&))/10
    IF CofGpoints(r&)<>0 THEN
        'found a system centre
        GRAPHIC ELLIPSE (CofGx(r&)/CofGpoints(r&)-BlueDotSize&, CofGy(r&)/CofGpoints(r&)-BlueDotSize&) - (CofGx(r&)/CofGpoints(r&)+BlueDotSize&, CofGy(r&)/CofGpoints(r&)+BlueDotSize&),%BLUE,%BLUE,0
    END IF
NEXT



GRAPHIC REDRAW
WAITKEY$

GRAPHIC SAVE "c:\stormoutput.bmp"

END FUNCTION

[Chris Holbrook]

Thanks for that Paul, fascinating stuff.

I need to do some work on the image before your app. can use it - at the moment all the birds vanish in the first pass! So I'm attempting to find and rank all the colours by "number of pixels of same colour in the image". Then starting with the lightest colour, turn all pixels of that colour white, and continue until the image starts to appear depleted of birds. I can't think how else to calibrate it.

[Gösta H. Lovgren-2]

The problem I see, Chris, after looking at the image blown a number of times is the starlings are not a single color. IOW, looking at individual pixels (image is 800x600) they are a range of (blue)greys. Some are only a couple pixels across and some are groups of px, 3 or 4 high x 6 or more across. Ans some are just single px.

Maybe a higher res picture. (?) Or a BW version instead of color (better contrasts(?))

Another thought is to maybe examine pxl and considering anything darker than RGB (152, 168, 193) (12,691,608) a starling as that *appears* (to me) the darkest sky color.

Sort of

Darkest_Sky_Color = RGB (152, 168, 193)
 
 
If px > Darkest_Sky_Color and Last_Px = < Darkest_Sky_Color& then incr starlings
 
Last_Px = This_Px

===================================
"When you have to kill a man,
it costs nothing to be polite."
Sir Winston Churchill (1874-1965)
===================================

Just a thought. Anyway a neat exercise to play with.

[Gösta H. Lovgren-2]

Just made a BW copy. Should be much easier to count (except for a few on the edges/corners). Only two colors.

Cheating on the example though.

======================================================
"I don't want to achieve immortality through my work;
I want to achieve immortality through not dying."
Woody Allen (1935-)
======================================================

[Chris Holbrook]

Just made a BW copy.

Gösta, Thanks for the BW copy - how did you make it?

[Eric Pearson]

remove sig

Chris --

In that picture, some birds pretty clearly ovlap so a larger single dark spot probably represents 2 or more birds. Some of the clumps are pretty large.

Why not simply count the number of dark pixels in the picture, then divide by your best guess of pixels-per-bird?

-- Eric

[Paul Dixon]

The brightness of the image varies a lot so it's difficult to set an overall threshold to say a point darker than the threshold is a bird and if it's lighter then it's sky.
Instead, you need to vary the threshold throughout the picture by looking, e.g. at a 100x100 pixel block and detecting birds in that block, move on a little and repeat in the next block which will have a different overall brightness and a different threshold. This will be quite a slow process.

The attached code uses a very crude approximation by just looking at 1 pixel either side and deciding if the change in brightness across those 3 pixels is high enough to suspect a bird. It does quite a good job and is quick but is not perfect.

EDIT: The one pixel either side method isn't as good as it first looked. Things will have to be done in blocks

Paul.

'PBCC5 program
#BREAK ON
#INCLUDE "win32api.inc"

%showscreenupdate=1

FUNCTION PBMAIN() AS LONG
REGISTER pix1 AS LONG
REGISTER y AS LONG

Storm$="c:\starlings2.bmp"


'get the size of the bitmap
nFile& = FREEFILE
OPEN Storm$ FOR BINARY AS nFile&
GET #nFile&, 19, nWidth&
GET #nFile&, 23, nHeight&
CLOSE nFile&

'load the bitmap of the storm system
GRAPHIC BITMAP LOAD Storm$,nWidth&,nHeight&,%BLACKONWHITE TO hBmp1???

GRAPHIC ATTACH hBmp1???,0
GRAPHIC GET BITS TO Image1$

xsize& = CVL(image1$,1)
ysize& = CVL(image1$,5)

'show it on screen so I can see the effect of each stage of processing
GRAPHIC WINDOW "Storm ID",1,1,xsize&,ysize& TO hWin1???
GRAPHIC ATTACH hWin1???,0 ,REDRAW

GRAPHIC COPY hBmp1???,0

DIM bmp(xsize&-1,ysize&-1) AS LONG AT STRPTR(Image1$)+8

PRINT xsize&,ysize&
PRINT LEN(image1$)

GRAPHIC REDRAW

CONSOLE SET LOC 800,0
CONSOLE SET FOCUS


PRINT "Original image"
WAITKEY$
PRINT "Improving contrast"

contrast##=0.95

       
'scan the image to find those points which are significantly darker than the points either side
FOR x& = 1 TO xsize&-1-1
    FOR y& = 0 TO ysize&-1
        pix1&=bmp(x&,y&)
        pix2&=bmp(x&-1,y&)
        pix3&=bmp(x&+1,y&)

        brightness1&= (pix1& AND 255) + ((pix1& AND &h0000ff00) \256 ) + ((pix1& AND &h00ff0000) \ 65536)
        brightness2&= (pix2& AND 255) + ((pix2& AND &h0000ff00) \256 ) + ((pix2& AND &h00ff0000) \ 65536)
        brightness3&= (pix3& AND 255) + ((pix3& AND &h0000ff00) \256 ) + ((pix3& AND &h00ff0000) \ 65536)

        'if pix1 is darker than those either side, identify it as a possible bird
        IF ((brightness2&/brightness1&) > contrast##) AND ((brightness3&/brightness1&) > contrast&) THEN
             GRAPHIC SET PIXEL (x&,y&),%BLACK
        ELSE
             GRAPHIC SET PIXEL (x&,y&),%WHITE
        END IF


    NEXT
#IF %DEF (%showscreenupdate)
    GRAPHIC REDRAW
#ENDIF

NEXT

'copy new image back to array
GRAPHIC GET BITS TO Image1$
REDIM bmp(xsize&-1,ysize&-1) AS LONG AT STRPTR(Image1$)+8
           


PRINT "Identifying individual starlings"
'identify each system from the others
NextColour&=1
DIM systemlocationx(100000) AS LONG
DIM systemlocationy(100000) AS LONG

FOR x& = 0 TO xsize&-1
    FOR y& = 0 TO ysize&-1
        GRAPHIC GET PIXEL (x&,y&) TO pix1&
        IF pix1& = %WHITE THEN
            GRAPHIC PAINT (x&,y&),NextColour&,%BLACK
            systemlocationx(NextColour&)=x&
            systemlocationy(NextColour&)=y&

            NextColour&=NextColour&+1
        END IF
    NEXT

#IF %DEF (%showscreenupdate)
    GRAPHIC REDRAW
#ENDIF

NEXT
GRAPHIC REDRAW

PRINT NextColour&-1;"Starlings found"
WAITKEY$

'colour them all white so they can be seen on the final picture
FOR x& = 0 TO xsize&-1
    FOR y& = 0 TO ysize&-1
        GRAPHIC GET PIXEL (x&,y&) TO pix1&
        IF pix1& <> %BLACK THEN
            GRAPHIC SET PIXEL (x&,y&),%WHITE

         END IF
    NEXT
NEXT
GRAPHIC REDRAW

WAITKEY$

GRAPHIC SAVE "c:\starlingoutput.bmp"

END FUNCTION

[Chris Holbrook]

Why not simply count the number of dark pixels in the picture, then divide by your best guess of pixels-per-bird?

Eric, that is a reasonable approach on an image-at-a-time basis. Pixels per bird brings in yet another variable, the distance between the camera and the flock. What I am hoping to work out is how to distinguish an irregular shape from its background. I'm sure that this has many applications and people are doing it all the time in process control, crop measurement, security, military applications. I would just like to join that club, even as a very junior member!

[Conrad Hoffman]

FWIW & AFAIK the biggest application of this is cell counting in medical applications. Google "cell counting" and software to get a good overview of what's out there. Hopefully there will be some papers that show how they go about it.

[Gösta H. Lovgren-2]

Gösta, Thanks for the BW copy - how did you make it?

Paint Brush Pro (a PhotoShp type program) has a BW option.

What I was going to do (haven't started yet) was look at 5x5 (or 4x or 3x) blocks of pixels and if any were not zero), count it as a bird. In BW, it should be pretty easy. In color, much harder.

i{later} Finished. I came up with 3,110 Birds in 12,065 Blocks.

Building on Paul's code'

 'PBWIN 9.00 - WinApi 05/2008 - XP Pro SP3
'http://www.powerbasic.com/support/pbforums/showthread.php?t=39425
'Original code by Paul Dixon
#Register All ' suggestion by Theo G. faster running?
'#Dim All                             
#Include "WIN32API.INC"
Macro wate
  Graphic ReDraw
  Graphic WAITKEY$
End Macro
'
Function PBMain
  Local Image1, Pic As String 
  Local Margin, ctr, ctr1, ctr2, ctr3, hctr, wctr, Bird_Count, Bird_Size, xWidth, xHeight, nWidth, nHeight, nfile As Long
  Local Block_Size, Blocks_Total As Long
  Local hBmp1, hWin1 As Dword
'  
 Pic$ = "starlings_BW.bmp"
 nfile = FreeFile
'
Open Pic$ For Binary As nFile& 'get size of bitmap
Get #nFile&, 19, nWidth 'width begins at 19th byte
Get #nFile&, 23, nHeight 'Height begibsa at 23rd byte
Close nFile&
'
'load the bitmap 
Graphic Bitmap Load Pic$, nWidth&, nHeight& To hBmp1 'assign the bitmap to a handle (DC?)
'
Graphic Attach hBmp1, 0 'assign a target for future graphic satements
Graphic Get Bits To Image1$ 'put the bmp ito a string
'
xWidth = Cvl(image1$,1) 'width is first 4 bytes (long)
xHeight = Cvl(image1$,5) 'Height is next four
'
'show it on screen so I can see the effect of each stage of processing
Graphic Window "Starlings", 1, 1, xWidth, xHeight To hWin1 'assign to a window handle
Graphic Attach hWin1, 0, ReDraw 'assign a target for future graphic satements
'
Graphic Copy hBmp1???,0 'copy Pic$ to "Starlings" window created above
'
Dim bmp(1 To xWidth, 1 To xHeight) As Long At StrPtr(Image1$) + 8 'create a working array
'
'
'no birds should be bigger than Bird_Size
  Bird_Size = 5
  Margin = 3 'Bird_Size \ 2
  Reset Bird_Count            
  Block_Size = Bird_Size * Bird_Size 
'  
          'start down a little   'stop up a little
   For hctr = Margin   To xHeight - Margin  Step Bird_Size
     For wctr = Margin To xWidth - Margin Step Bird_Size 
         Incr Blocks_Total
         'now see if any filled
         Reset ctr
         For ctr1 = -2 To 2
           For ctr2 = -2 To 2
              If bmp(wctr + ctr2, hctr + ctr1) < 1 Then Incr ctr '> 0 is black color
           Next ctr2
         Next ctr1
'         
         If ctr > 0 And ctr < Block_Size Then ' Case of block being all black
            Incr Bird_Count
         End If   
     Next wctr
   Next hctr
'
'   
'
Graphic ReDraw
'
  wate
'
   Open "Check.txt" For Output As #nfile
   Print #nfile, Using$("#, Wide  #, High", xwidth, xHeight)   
   Print #nfile, Using$("#, Birds in #, Blocks ", Bird_Count, Blocks_Total)
'                                                                       
   For hctr = 1 To xHeight        
    Print #nfile, Using$("#,#### ", hctr);
     For wctr = 1 To 10 'just to see how it looks
       Print #nfile, Using$("#, ", bmp(wctr, hctr));
     Next wctr
     Print #nfile, " "
   Next hctr
   Close
End Function
'

In file

640 Wide 480 High
3,110 Birds in 12,065 Blocks
...

While probably not an accurate count, I'd bet it's pretty close.

[David Roberts]

Gösta's tip re grey scale is the way to go with this image but I reckon you've been a bit heavy on the contrast peddle, Gösta.

I've used less contrast on the whole image then applied selective contrast to the lower part. Paint Shop Pro has an Edge Effect and the Trace Contour works a treat on the image so far giving a higher retention of them thar birds.



I tried your code Gösta but it GPF'd and I haven't had time to check why.

[Gösta H. Lovgren-2]

Gösta's tip re grey scale is the way to go with this image but I reckon you've been a bit heavy on the contrast peddle, Gösta.

A heavy foot has been a problem for me all my life {sigh}.

I've used less contrast on the whole image then applied selective contrast to the lower part. Paint Shop Pro has an Edge Effect and the Trace Contour works a treat on the image so far giving a higher retention of them thar birds.

I have no idea how to do that. Used PBP since it was version 1 and shareware. Just kept updating as each version came to support the original author (forget his name) who gave great support. I should have stopped around version 4 or 5 (my last is 8.xx) because then it started becoming more and more complex for the (very) casual user. {oh well}.

I tried your code Gösta but it GPF'd and I haven't had time to check why.

It runs fine here (picture should be in same folder as program). You may have to put in "Drve:\Folder\Starlings..." if you have it elsewhere (though I don't think that would cause a GPF).

===================================
"The gods too are fond of a joke."
Aristotle (384-322 B.C.)
===================================

[Gösta H. Lovgren-2]

Black & White estimate
3,110 Birds in 12,065 Blocks of 5 square
Color Total = 4,819,473,112,545 Color Avg = 15,688,389

Color Estimate at 16,777,215 for a sky color
3,078 Birds in 3,078 Pixels (at 1 Px per bird) 1% of total

Grey Estimate at 16,777,215 for a sky color
3,078 Birds in 8,839 Pixels (at 1 Px per bird) 3% of total

'*****************

My best guess is there's about 3,088.666666666666666667 birds in the picture {grin}.


'

 'PBWIN 9.00 - WinApi 05/2008 - XP Pro SP3
'http://www.powerbasic.com/support/pbforums/showthread.php?t=39425
'Original code by Paul Dixon
#Register All ' suggestion by Theo G. faster running?
'#Dim All                             
#Include "WIN32API.INC"
Macro wate
  Graphic ReDraw
  Graphic WAITKEY$
End Macro
'
Macro pf = Print #nfile, 
'
Macro Load_Picture
  Open Pic$ For Binary As nFile& 'get size of bitmap
  Get #nFile&, 19, nWidth 'width begins at 19th byte
  Get #nFile&, 23, nHeight 'Height begibsa at 23rd byte
  Close nFile&
  '
  'load the bitmap 
  Graphic Bitmap Load Pic$, nWidth&, nHeight& To hBmp1 'assign the bitmap to a handle (DC?)
  '
  Graphic Attach hBmp1, 0 'assign a target for future graphic satements
  '
  Graphic Get Bits To Image1$ 'put the bmp ito a string
  '
  xWidth = Cvl(image1$,1) 'width is first 4 bytes (long)
  xHeight = Cvl(image1$,5) 'Height is next four
  '
  'show it on screen so I can see the effect of each stage of processing
  Graphic Window "Starlings", 1, 1, xWidth, xHeight To hWin1 'assign to a window handle
  Graphic Attach hWin1, 0, ReDraw 'assign a target for future graphic satements
  '
  Graphic Copy hBmp1,0 'copy Pic$ to "Starlings" window created above
  '
  ReDim bmp(1 To xWidth, 1 To xHeight) As Long At StrPtr(Image1$) + 8 'create a working array
  ReDim Colrs(1 To xWidth * xHeight) 'put color values in
  '
  Graphic ReDraw
  wate
End Macro
'
Function PBMain
  Local Margin, ctr, ctr1, ctr2, ctr3, hctr, wctr, Bird_Count, Bird_Size, xWidth, xHeight, nWidth, nHeight, nfile As Long
  Local Bird_Pixels_DR, Pixels_Total, Bird_Pixels, Bird_Color, Color_Avg, Colrs(), Block_Size, Blocks_Total As Long
  Local Image1, Pic As String 
  Local hBmp1, hWin1 As Dword
  Local Color_Ttl As Quad
'  
 Pic$ = "starlings_BW.bmp"
 nfile = FreeFile 
'
   Load_Picture
'
 
   For hctr = 1 To xheight
     For wctr = 1 To xWidth     
       Colrs(xWidth * xHeight) = bmp(wctr, hctr)
       Color_Ttl = Color_Ttl + bmp(wctr, hctr)
     Next wctr
   Next hctr                                   
'                                               
  Color_Avg = Color_Ttl \ (xWidth * xHeight)
'
'no birds should be bigger than Bird_Size
  Bird_Size = 5
  Margin = 3 'Bird_Size \ 2
  Reset Bird_Count            
  Block_Size = Bird_Size * Bird_Size 
'
' Any color within a Bird_Size Block = a bird 
          'start down a little   'stop up a little
   For hctr = Margin   To xHeight - Margin  Step Bird_Size
     For wctr = Margin To xWidth - Margin Step Bird_Size 
         Incr Blocks_Total
         'now see if any filled
         Reset ctr
         For ctr1 = -2 To 2
           For ctr2 = -2 To 2
              If bmp(wctr + ctr2, hctr + ctr1) < 1 Then Incr ctr
           Next ctr2
         Next ctr1
'         
         If ctr > 0 And ctr < Block_Size Then ' Case of block being all black
            Incr Bird_Count
         End If   
     Next wctr
   Next hctr
'
'>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
'<<< Now do it in color
'
'   
   Bird_Color = RGB(90, 100, 100) 'picked from photo as darkest sky color
                                   'Anything darker (lower number) is considered a part of a bird  
 Pic$ = "starlings.bmp"
 nfile = FreeFile 
   Load_Picture
 
 
   For hctr = 1 To xHeight
     For wctr = 1 To xWidth
         Incr Pixels_Total
         'now see if any filled
          If bmp(wctr, hctr) <  Bird_Color Then Incr Bird_Pixels
     Next wctr
   Next hctr
'
'>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
'<<< Now do it in DR's contrasted
'
'   
   Bird_Color = RGB(255, 255, 255) 'White
                                   'Anything darker (lower number) is considered a part of a bird  
 Pic$ = "starlingsDR.bmp"
 nfile = FreeFile 
   Load_Picture
 
 
   For hctr = 1 To xHeight
     For wctr = 1 To xWidth
         'now see if any filled
          If bmp(wctr, hctr) <  Bird_Color Then Incr Bird_Pixels_DR
     Next wctr
   Next hctr
'
'
'
  wate
'
   Open "Check.txt" For Output As #nfile
   Print #nfile, Using$("#, Wide  #, High = #, Total = #,  #,", xwidth, xHeight, xwidth * xHeight, Pixels_Total)   
   Print #nfile, ,"Black & White estimate"
   Print #nfile, Using$("#, Birds in #, Blocks of 5 square", Bird_Count, Blocks_Total)
   Print #nfile, Using$("Color Total = #,  Color Avg = #,", Color_Ttl, Color_Avg)
   ctr = 1
   pf " "                                                                    
   pf Using$("Color Estimate at #, for a sky color", Bird_Color)
   pf Using$("#, Birds in #, Pixels (at # Px per bird) #% of total", _
              Bird_Pixels \ ctr, Bird_Pixels, ctr,  Bird_Pixels / Pixels_Total * 100)
 
   ctr = 1
   pf " "                                                                    
   pf Using$("Grey Estimate at #, for a sky color", Bird_Color)
   pf Using$("#, Birds in #, Pixels (at # Px per bird) #% of total", _
              Bird_Pixels \ ctr, Bird_Pixels_DR, ctr,  Bird_Pixels_DR / Pixels_Total * 100)
 
   Close
End Function
'

=====================================================
"Without accepting the fact that everything changes,
we cannot find perfect composure.
But unfortunately, although it is true,
it is difficult for us to accept it.
Because we cannot accept
the truth of transience,
we suffer."
Shunryu Suzuki
=====================================================

[Chris Holbrook]

My best guess is there's about 3,088.666666666666666667 birds in the picture {grin}.

That is the equivalent of 3,085 starlings, one seagull and a brace of sparrows.

I am profoundly impressed.

[Gösta H. Lovgren-2]

That is the equivalent of 3,085 starlings, one seagull and a brace of sparrows.

I am profoundly impressed.

Why? That I caught the sparrows?

Actually Paul's code showing how to put a graphic into an array:

Graphic Get Bits ...
ReDim bmp(1 To xWidth, 1 To xHeight) As Long At StrPtr(Image1$) + 8 ...

made it absurdly easy, which is not to say accurate. Actually I think the two sparrows may be seagull turds (they are so small details are hird to make out.)

===========================================
"A pint of sweat saves a gallon of blood."
General George S. Patton (1885-1945)
===========================================

[Cliff Nichols]

Quote:
Originally Posted by Gösta H. Lovgren-2
My best guess is there's about 3,088.666666666666666667 birds in the picture {grin}.

That is the equivalent of 3,085 starlings, one seagull and a brace of sparrows.

I am profoundly impressed.

All together now....

....Annnnd a partridge in a pear treeeeeee!!!!!



Seriously though I think the problem is more a context of color range vs the pixel being looked at to determine results

[Chris Holbrook]

Seriously though I think the problem is more a context of color range vs the pixel being looked at to determine results

Yes. If you were to use such a technique in "production", you might have to use different strategies to maximise the number of objects discerned, assuming that the birds are smaller than cloud features.Geese against a "mackerel sky" would be challenging.

[Gösta H. Lovgren-2]

Yes. If you were to use such a technique in "production", you might have to use different strategies to maximise the number of objects discerned, assuming that the birds are smaller than cloud features.Geese against a "mackerel sky" would be challenging.

Assuming the picture was in color, It seems geese would be easier and more accurate in some respects, if one could find a color unique to the geese (say a green or red or brown). Then one could just search for that particular color (grouping). In the first run I examine blocks of 25. If one were searching for Red, then any reds in the block would be a (1) goose.

What made this particular exercise rewarding for me was the technique Paul showed (putting the image into an array of pixel colors). Quite clever really for one unfamilar with graphics work. (me that is, not Paul)

At any rate, here's the colors for a 21H x 11W section which looks like it has 6 of the smaller (fainter) birds.

 
section.bmp
  1  8,754,606  8,688,813  8,754,093  8,688,300  8,688,300  8,688,300  8,819,372  8,885,165  8,885,165  8,885,165  8,885,165 
  2  8,885,163  8,885,163  8,754,347  8,688,554  8,688,300  8,688,300  8,688,300  8,688,300  8,819,372  8,885,165  8,884,650 
  3  8,819,370  8,819,370  8,688,554  8,688,554  8,688,554  8,688,554  8,688,300  8,688,300  8,819,372  8,885,165  8,884,650 
  4  8,819,370  8,819,370  8,754,347  8,688,554  8,688,554  8,688,554  8,688,300  8,688,300  8,885,165  8,885,165  8,950,445 
  5  8,885,163  8,885,163  8,885,163  8,885,163  8,885,163  8,885,163  8,819,372  8,885,165  8,885,165  8,885,165  8,885,165 
  6  8,885,163  8,885,163  8,885,163  8,885,163  8,885,163  8,885,163  8,819,372  8,885,165  8,885,165  8,819,372  8,819,372 
  7  8,885,163  8,885,163  8,950,956  8,885,163  8,885,163  8,885,163  8,819,372  8,885,165  8,885,165  8,819,372  8,819,372 
  8  8,885,165  8,885,165  8,950,956  8,885,163  8,819,370  8,819,370  8,819,372  8,885,165  8,885,165  8,819,372  8,819,372 
  9  8,885,165  8,950,958  8,950,956  8,885,163  8,819,370  8,819,370  8,885,163  8,885,163  8,885,165  8,819,372  8,885,165 
 10  9,016,749  8,885,163  9,016,490  8,884,904  8,950,443  8,950,443  8,950,445  8,950,445  9,016,238  8,950,445  8,885,165 
 11  9,082,542  8,753,577  8,490,146  8,161,181  8,621,478  9,082,029  9,082,031  9,016,238  9,016,238  8,950,445  8,819,372 
 12  9,082,542  8,819,370  8,424,099  7,502,997  7,371,411  7,568,790  8,753,066  8,950,445  9,016,238  9,016,238  9,016,238 
 13  8,885,163  9,016,749  9,148,335  8,753,577  8,818,857  8,950,443  8,884,652  8,950,445  8,818,859  8,884,652  9,082,031 
 14  9,016,749  8,885,163  8,950,956  8,819,370  9,016,749  9,148,335  8,885,165  8,621,993  8,029,343  8,226,722  8,884,652 
 15  9,148,335  8,950,956  9,016,749  8,885,163  8,950,956  8,819,370  8,753,579  8,687,786  8,029,856  8,227,235  8,884,652 
 16  9,082,029  8,884,650  8,950,956  8,950,956  9,082,542  8,885,163  9,016,751  9,148,337  8,555,945  8,687,531  9,147,824 
 17  9,147,822  9,016,236  9,016,749  8,885,163  8,950,956  8,950,956  8,687,786  8,621,993  7,898,015  8,292,773  9,082,031 
 18  8,950,958  8,885,165  9,016,751  8,950,958  8,819,372  8,950,958  8,885,165  8,819,372  7,964,063  8,490,407  9,147,569 
 19  8,950,958  8,885,165  9,016,751  8,950,958  8,885,165  9,016,751  8,885,165  9,082,544  8,687,786  8,950,958  9,147,824 
 20  9,016,238  9,016,238  9,082,544  9,016,751  8,950,958  9,016,751  8,885,165  9,082,544  9,082,544  9,082,544  9,016,751 
 21  9,016,496  9,082,289  9,082,544  8,950,958  8,950,958  9,016,751  9,016,751  8,950,958  9,016,751  9,016,751  8,950,958

The .bmp is attached and here is what it looks like blown up 300 times:
http://www.SwedesDock.com/powerbasic/section21x11.bmp (size exceeds forum upload limit)

Just kinda neat.

======================================================
"How can I lose to such an idiot?"
A shout from chessmaster Aaron Nimzovich (1886-1935)
======================================================

Oh, just for the record, I made no claims as to the accuracy of the number of birds, only "My best guess is there's about 3,088.666666666666666667 birds in the picture {grin}."

[Paul Dixon]

Gösta,

if one could find a color unique to the geese

It's not that simple
First, if a bird almost fills a single pixel then that pixel will be a dark shade of bird but if that same bird is half way between 2 pixels then each pixel will be a mix of the background colour and the bird so the colour of the pixel no longer identifies that there is a bird there. It might be different if the picture was a close up of the birds where each bird occupied many pixels so some pixels were always full of bird.

Second, the images are JPGs. If you look at your zoomed in shot you can see lots of artifacts introduced by the JPG compression. These show up as patches of varying light and dark pixels around each bird. The APPARENT colour to a person might be a reasonable match for the real image but to a computer these colours are just noise and crude approximations to the real colour in the scene and don't match at all. This is the reason my simple method earlier of checking the brightness each side of a pixel didn't work, it identified the JPG artifacts as changes in colour so thought they were birds.

I think you're on the right lines to zoom in and take a look at what a computer sees when it looks at the image pixel by pixel. You just need to come up with the algoritm that the computer needs to correctly identify bird from noise. Colour alone is never going to work, you need to look at the difference in colour between pixels in small areas of the image.


Paul.