Search similitary of a picture in folder of thousand others

[Kwai chang caine]

@Idle
Waoooouuuhhh !!!
Thanks a lot my friend IDLE, i test your code on my thousand records and say to you the result
Have a very good end of years

For that particular image, I think OCR could do it.

Thanks IdeasVacuum, it's also an idea..
But i'm not sure, an OCR is easy to use for searching a picture in thousands other
Or perhaps i not know all the functions of OCR

[Zach]

OCR = optical character recognition.

Basically, it trains the computer to recognize certain shapes as letters and numbers. I use it all the time on Blu-Ray subtitle files, to generate text-based output.

You could use it like that, to look for album/artists names.

[Kwai chang caine]

Aaaaah you want to say, try to recognize the title on the records, i'm not sure it's simple for the OCR
Because all the title have text drawing with much things around, too much color, style, etc... and not really writing...very difficult to recognize
A little bit like a captcha

I don't know what is the time for recognize one record in several thousand...
And not sure, it's not better in time, to try to do that with my eyes

But it's an idea.... some OCR are very powerfull....
Thanks for your answer

[Kwai chang caine]

I have also try with OpenCv, but for the moment, the result is not really "decisive" (It's the less i can say )
And after several day of research, try several methods, freewares, sharewares, shitware , etc ...

I have found 3 ways for the moment to found similar picture:
Image dupeles v1.6.3 works good but shareware
Visipics v1.31 Fr Free works very good, but impossible to take the name of the image found
ImageMagick Free and for the moment it's the only one, i can use with PB (Thanks to DANILO DANILO/KCC = The dream team )

So ATTENTION !!!
After the PQ-Code, the most quick code to compare in the world
KCC make the CP-Code (CowPat-Code), the most slow code to compare in the universe
See the followed link
That's works, even if the picture is a little bit rotated, and more lightning
So in this case (The record saragoband, big bamboo), follow my instructions :

1/ Run the application
2/ Search a wife
3/ Married with her
4/ Make three childrens
5/ Return to your PC...normally the picture is found

So if the picture is not rotate (the case of the other) the code is more quick
The unity of KCC is not the millisecond, the second,................ but more the minut, see perhaps...the hours

If someone can make something more fast don't hésitate i'm not offensed

[wilbert]

KCC make the CP-Code (CowPat-Code), the most slow code to compare in the universe
https://files.cloud.orange.fr/cloudUpDo ... Error=true

The link doesn't work for me.
Some message about a wassup cookie that is missing in the request.

[Kwai chang caine]

Excuse me.....
This is the good link
http://erdsjb.free.fr/PureStorage/Provi ... cords2.zip

[BasicallyPure]

KCC I think this will be of interest to you.
http://phash.org/

BP

[Kwai chang caine]

Hello BASICALLYPURE
Since all this time i search a solution, i have also found pHash.
But the problem is i have not your knowledge
I don't know the C, it's not an EXE, and nobody already have created a wrapper or an interface in PB like JHPJHP for opencv

You have right ,apparently it's exactely that i need, because i have tested the site research and the result is verygood.
So for the time, i continue with my interface of ImageMagick, that's works not bad, with again little error, but it's better than nothing.
Like i'm a waffer, i have learning to not be difficult

The true problem of my code, is for the moment i have not found how create a number, a key,like md5 but with an analyse of the image, for each image, a little bit like your splendid works of histogram, in fact i have found nice, but not understand all
My goal is create a style of key for each records and put this key in database for have a quick search.
For the time, my code is forced to compare each 12000 images with each record searched.
Imagine yesterday, for found 50 records my code have take 10 hours, it's not again the level of google image...
For have a more quick result, i have reduce the 12000 record to 100x100 and imagemagick found more quickly.
Thanks again for your interest for my lost case

[wilbert]

This might be some interesting posts about the topic ...
http://www.hackerfactor.com/blog/?/arch ... -That.html
http://www.hackerfactor.com/blog/?/arch ... ke-It.html

[Kwai chang caine]

Thanks WILBERT but during my numerous research i have also see this blog.
But it's chinese for kcc
In fact, i believe to have read all the web, before disturb you, and have eyes like a chameleon
Much explanation. further in english, and no exe, and link on c
Also in python, i not know better , further broken

I have also, since this time, found this link
http://www.pureftpd.org/project/libpuzzle
First i believe found the solution, but same problem...not understand how use it

I have found a way for calculate the primary color of all a picture with imagemagick.
I hope with that, sort my records by color, like that not be forced to compare 12000 records each time.

It's for the moment, the only thing i have found...

[wilbert]

The dHash from the first link I posted seems pretty easy to implement.
pHash seems a little more complicated since I'm not familiar with DCT.

Edit:
See also http://www.purebasic.fr/english/viewtop ... 35#p436035 for my npHash module.

[Kwai chang caine]

The dHash from the first link I posted seems pretty easy to implement.

Since 8 years i try to programming in PB, i have destroy in my head the word "EASY"
The only thing is easy for me all day...it's run PB IDE (And again ... )
After for me it's a fight of all the days


By "easy" you mean this link for use it with PHP ?
http://jax-work-archive.blogspot.com/20 ... dhash.html

[wilbert]

This should be dHash (hope I didn't make any mistakes)

Code: Select all

UseJPEGImageDecoder()
UsePNGImageDecoder()

Procedure.q dHash(filename.s)
  Protected.i img, x, y, c, l0, l1, b, hash.q
  
  img = LoadImage(#PB_Any, filename)
  If IsImage(img)
    ResizeImage(img, 9, 8)
    StartDrawing(ImageOutput(img))
    For y = 0 To 7
      c = Point(0, y)
      l0 = (c >> 8 & $ff) << 2 + (c & $ff) << 1 + (c >> 16 & $ff) 
      For x = 1 To 8
        c = Point(x, y)
        l1 = (c >> 8 & $ff) << 2 + (c & $ff) << 1 + (c >> 16 & $ff)
        If l0 < l1 : hash | 1 << b : EndIf
        l0 = l1
        b + 1
      Next
    Next
    StopDrawing()
    FreeImage(img)
  EndIf
  
  ProcedureReturn hash
EndProcedure

Procedure.i dHashCompare(hash1.q, hash2.q)
  Protected.i result, b
  hash1 ! hash2
  For b = 0 To 63
    If hash1 & (1 << b)
      result + 1
    EndIf
  Next
  ProcedureReturn result
EndProcedure

The dHash procedure returns a 64 bit hash value.
If you compare two hashes using dHashCompare the lower the value, the better the match.
A value of 0 most likely is the same image, a value > 10 is most likely a different image.

[Kwai chang caine]

Waooooouuuh i don't believe my eyes !!!!

I have try your crazy code, and the result is enough good
I have not testing in all the case (Rotate, size, lightning, etc...)
But for the moment, it's good !!!! see yourself

Compare two differents picture of extremely different record
-2770167811735213228
2609421560056017481
34

Compare exactely the same picture of the same record
-2770167811735213228
-2770167811735213228
0

Compare two differents picture of the same record
-2770167811735213228
-2842226509793678442
13

Compare the same picture rotating
-2770167811735213228
9173101322984576773
21

Compare the same picture lightning
-2770167811735213228
-2770167811735213100
1

It's a miracle !!!!
http://erdsjb.free.fr/PureStorage/Provi ... ilbert.zip


One thousand of thanks WILBERT !!!

[wilbert]

I'm glad it works for you

You could try a minor modification

Code: Select all

UseJPEGImageDecoder()
UsePNGImageDecoder()

Procedure.q dHash(filename.s)
  Protected.i img, x, y, c, l0, l1, b, hash.q
  
  img = LoadImage(#PB_Any, filename)
  If IsImage(img)
    ResizeImage(img, 9, 8)
    StartDrawing(ImageOutput(img))
    For y = 0 To 7
      c = Point(0, y)
      l0 = (c >> 8 & $ff) << 2 + (c & $ff) << 1 + (c >> 16 & $ff) 
      For x = 1 To 8
        c = Point(x, y)
        l1 = (c >> 8 & $ff) << 2 + (c & $ff) << 1 + (c >> 16 & $ff)
        If l0 < l1 : hash | 1 << b : EndIf
        l0 = l1
        b + 1
      Next
    Next
    StopDrawing()
    FreeImage(img)
  EndIf
  
  ProcedureReturn hash
EndProcedure

Procedure.q dHash2(filename.s)
  Protected.i img, x, y, c, l0, l1, b, hash.q
  
  img = LoadImage(#PB_Any, filename)
  If IsImage(img)
    ResizeImage(img, 11, 10)
    StartDrawing(ImageOutput(img))
    For y = 1 To 8
      c = Point(1, y)
      l0 = (c >> 8 & $ff) << 2 + (c & $ff) << 1 + (c >> 16 & $ff) 
      For x = 2 To 9
        c = Point(x, y)
        l1 = (c >> 8 & $ff) << 2 + (c & $ff) << 1 + (c >> 16 & $ff)
        If l0 - l1 < 8 : hash | 1 << b : EndIf
        l0 = l1
        b + 1
      Next
    Next
    StopDrawing()
    FreeImage(img)
  EndIf
  
  ProcedureReturn hash
EndProcedure

Procedure.i dHashCompare(hash1.q, hash2.q)
  Protected.i result, b
  hash1 ! hash2
  For b = 0 To 63
    If hash1 & (1 << b)
      result + 1
    EndIf
  Next
  ProcedureReturn result
EndProcedure

Does dHash2 give you better results ?
It's a small modification that discards the outer parts.