How to detection similarity of two Image

dear all members
I am trying to find rate of difference or rate of similarity of two
Images.
anyone can give me any Idea ?
for example :
if we have 2 Image : im1.jpg and im2.jpg
----------------------------------
f1=imread('im1.jpg');
f2=imread('im2.jpg');
F1t2(f1);
F22(f2);
similarity(f1,f2)=?
-------------------------------
How can I compare this two Fouriers?
How can I find rate of similarity of this two image?
the expected results are something like this:
similarity(f1,f2).9
similarity(f1,f3).7
similarity(f1,f4)4.5
....
Thanks Start a New Thread

perhaps you may try image subtraction of image difference operations...if the images are of the same size then these image calculations may determine the diferences you are after...hope this helps
cheers

--- On Thu, 7/1/10, mohammad wrote:
From: mohammad
Subject: [imagedsp] How to detection similarity of two Image
To: i...
Received: Thursday, 7 January, 2010, 5:44 AM
dear all members
I am trying to find rate of difference or rate of similarity of two
Images.
anyone can give me any Idea ?
for example :
if we have 2 Image : im1.jpg and im2.jpg
----------------------------------
f1=imread('im1.jpg');
f2=imread('im2.jpg');
F1=fft2(f1);
F2=ff2(f2);
similarity(f1,f2)=?
-------------------------------
How can I compare this two Fouriers?
How can I find rate of similarity of this two image?
the expected results are something like this:
similarity(f1,f2)=99.9
similarity(f1,f3)=98.7
similarity(f1,f4)=34.5
....
Thanks

_____________________________________ Start a New Thread

You can also compute the correlation between these two images and look for the picks; the same which we apply in template matchings and so forth.

bests Start a New Thread

Hello mr. mohammad
for similarity evaluation
you can calculate the corrolation coeficient of two images
using this command: corr2(f1,f2)
this will give you anumber in the interval of [0 1]
1 shows full similarity and 0 shows no similarity.
you also can comper the histogram of images
or simply the squar error between to images as :
Error=sum(sum((f1-f2).^2)) Start a New Thread

Dear, Mehrnaz
Thanks for ur Idea.
I tried both of the your seggustion ( Corr2( ) , and Sum(sum( )).
but these method do the similarity estimation with pixel-by-pixel. and they are not robust against a bit modification .
I need an effective Discriminative Invariant Features of Image that be robust against small modification for example rotation.
Thanks again ,
plz Give me your Idea , If you have another one.
In my opinion , there is need to robust feature in frequency domain !
for example if we have 3 blocks:
A1248 253 251 253
245 252 250 253
245 252 251 254
248 252 252 254

A2252 252 251 251
252 252 251 251
252 252 252 252
253 252 252 252

D1229 229 229 229
229 229 229 229
229 0 229 229
229 229 229 229

D21 200 3 0
80 2 30 50
100 50 200 90
0 2 70 1

we have to be able to find percentage of similarity of these blocks.
in this example -
the result has to be something like these:
1- similarity(A1,A2) -----> 99.99%
2- similarity(A1,D1) -----> 80%
3- similarity( A1,D2) --------> 10%

Best Regards

Mohammad Akbarpour Sekeh
Student of PhD in Computer Forensic -IFD area
Faculty of Computer Science and Information system
University Technology Malaysia
81310 UTM Skudai, Johor, Malaysia
Cell-Phone:+60-127890645

________________________________
From: mehrnaz fani
To: mohammad
Cc: i...
Sent: Fri, 8 January, 2010 14:12:51
Subject: Re: [imagedsp] How to detection similarity of two Image
Hello mr. mohammad
for similarity evaluation
you can calculate the corrolation coeficient of two images
using this command: corr2(f1,f2)
this will give you anumber in the interval of [0 1]
1 shows full similarity and 0 shows no similarity.
you also can comper the histogram of images
or simply the squar error between to images as :
Error=sum(sum((f1-f2).^2)) Start a New Thread

Try xcorr2(), then… It gives a peak at a location depending on the amount of a shifting perturbation. If you have rotative or scaling perturbations, then you may try multi resolution approaches together with KLT (for alignment). If you have scaling or rotation, so does the frequency domain. Please let us know if you find a robust feature there.



Best,



Omer



From: i... [mailto:i...] On Behalf Of mohammad akbarpour
Sent: Friday, January 08, 2010 9:30 AM
To: mehrnaz fani
Cc: i...
Subject: Re: [imagedsp] How to detection similarity of two Image



Dear, Mehrnaz
Thanks for ur Idea.
I tried both of the your seggustion ( Corr2( ) , and Sum(sum( )).
but these method do the similarity estimation with pixel-by-pixel. and they are not robust against a bit modification .
I need an effective Discriminative Invariant Features of Image that be robust against small modification for example rotation.
Thanks again ,
plz Give me your Idea , If you have another one.
In my opinion , there is need to robust feature in frequency domain !
for example if we have 3 blocks:
A1=
248 253 251 253
245 252 250 253
245 252 251 254
248 252 252 254

A2=
252 252 251 251
252 252 251 251
252 252 252 252
253 252 252 252

D1=
229 229 229 229
229 229 229 229
229 0 229 229
229 229 229 229

D2=
1 200 3 0
80 2 30 50
100 50 200 90
0 2 70 1

we have to be able to find percentage of similarity of these blocks.
in this example -
the result has to be something like these:
1- similarity(A1,A2) -----> 99.99%
2- similarity(A1,D1) -----> 80%
3- similarity( A1,D2) --------> 10%


Best Regards

Mohammad Akbarpour Sekeh
Student of PhD in Computer Forensic -IFD area
Faculty of Computer Science and Information system
University Technology Malaysia
81310 UTM Skudai, Johor, Malaysia
Cell-Phone:+60-127890645





_____

From: mehrnaz fani
To: mohammad
Cc: i...
Sent: Fri, 8 January, 2010 14:12:51
Subject: Re: [imagedsp] How to detection similarity of two Image

Hello mr. mohammad

for similarity evaluation

you can calculate the corrolation coeficient of two images

using this command: corr2(f1,f2)

this will give you anumber in the interval of [0 1]

1 shows full similarity and 0 shows no similarity.

you also can comper the histogram of images

or simply the squar error between to images as :

Error=sum(sum((f1-f2).^2))

_____________________________________ Start a New Thread

Dear All,

Hello and Happy new year!

For comapring two images or signals and for measuring rthe similarities, one of the approaches is to extract features and then, compare these features. As you mentioned these features are suppoosed to be invariant with respect to tranlation, rotation, and scale. For example co-occurrence matrix feature extraction method and Hu moments are widely used for image analysis and classification. However, selecting the aproach really depends on the application. You can find more about them in image processing and pattern recognition text books.

I hope this would be helpful.

Best regards,
Alireza
 

________________________________
From: mohammad akbarpour
To: mehrnaz fani
Cc: i...
Sent: Fri, January 8, 2010 8:30:29 AM
Subject: Re: [imagedsp] How to detection similarity of two Image

 
Dear, Mehrnaz
Thanks for ur Idea.
I tried both of the your seggustion ( Corr2( ) , and Sum(sum( )).
but these method do the similarity estimation with pixel-by-pixel. and they are not robust against a bit modification .
I need an effective Discriminative Invariant Features of Image that be robust against small modification for example rotation.
Thanks again ,
plz Give me your Idea , If you have another one.
In my opinion , there is need to robust feature in frequency domain !
for example if we have 3 blocks:
A1=
248     253     251     253
245     252     250     253
245     252     251     254
248     252     252     254

A2=
252     252     251     251
252     252     251     251
252     252     252     252
253     252     252     252

D1=
229     229     229     229
229     229     229     229
229     0         229     229
229     229     229     229

D2=
1        200       3      0
80      2          30     50
100    50        200   90
0        2          70     1
we have to be able to find percentage of similarity of these blocks.
in this example -
the result has to be  something like these:
1- similarity(A1, A2)   -----> 99.99%
2- similarity(A1, D1)   ----->  80%
3- similarity( A1,D2) --------> 10%

Best Regards

Mohammad Akbarpour Sekeh
Student of PhD in Computer Forensic -IFD area
Faculty of Computer Science and Information system
University Technology Malaysia
81310 UTM Skudai, Johor, Malaysia
Cell-Phone:+ 60-127890645

________________________________
From: mehrnaz fani
To: mohammad
Cc: imagedsp@yahoogroup s.com
Sent: Fri, 8 January, 2010 14:12:51
Subject: Re: [imagedsp] How to detection similarity of two Image
Hello mr. mohammad
for similarity evaluation
you can calculate the corrolation coeficient of two images
using this command: corr2(f1,f2)
this will give you anumber in the interval of [0 1]
1 shows full similarity and 0 shows no similarity.
you also can comper the histogram of images
or simply the squar error between to images as :
Error=sum(sum( (f1-f2).^ 2))



_____________________________________ Start a New Thread

Hi
For measuring similarity of two image you have to use image siedility metrics such as MSE,PSNR,SSIM,NC,...
Regards

dear all members
>I am trying to find rate of difference or rate of similarity of two
>Images.
>anyone can give me any Idea ?
>for example :
>if we have 2 Image : im1.jpg and im2.jpg
>----------------------------------
>f1=imread('im1.jpg');
>f2=imread('im2.jpg');
>F1t2(f1);
>F22(f2);
>similarity(f1,f2)=?
>-------------------------------
>How can I compare this two Fouriers?
>How can I find rate of similarity of this two image?
>the expected results are something like this:
>similarity(f1,f2).9
>similarity(f1,f3).7
>similarity(f1,f4)4.5
>....
>Thanks Start a New Thread

> try this paper and find more in its reference section.
> http://www.cs.uwa.edu.au/~gian/docs/hons_paper.pdf

Here is updated URL:

http://www.csse.uwa.edu.au/~pk/studentprojects/gian/hons_paper.pdf

Best regards,
Mateusz Loskot Start a New Thread