How to view eigenfaces, Re: Re: Face recognition using eigenfaces

Hi,

How would I acutally normalise the Eigenfaces so that I can view 
them?

Thanks,
Kirk

--- In OpenCV@xxxxxxxxxxxxxxx, "Salman Aslam" <salman.m.aslam@...> 
wrote:
>
> thankyou Parameswaran.
> 
> regards
> 
>   ----- Original Message ----- 
>   From: Parameswaran 
>   To: OpenCV@xxxxxxxxxxxxxxx 
>   Sent: Wednesday, March 29, 2006 11:34 AM
>   Subject: How to view eigenfaces, Re: [OpenCV] Re: Face 
recognition using eigenfaces
> 
> 
>   If you normalize the eigenface pixel values between 0 and 255, 
you can
>   view the image using the HighGUI window.
> 
>   Regards, :D
>   Paramesh.
> 
>   --- In OpenCV@xxxxxxxxxxxxxxx, "Salman Aslam" <salman.m.aslam@> 
wrote:
>   >
>   > is there a way to view the eigenfaces?
>   > 
>   > 
>   >   ----- Original Message ----- 
>   >   From: Parameswaran 
>   >   To: OpenCV@xxxxxxxxxxxxxxx 
>   >   Sent: Friday, March 17, 2006 9:10 PM
>   >   Subject: [OpenCV] Re: Face recognition using eigenfaces
>   > 
>   > 
>   >   You cant save float eigenfaces..
>   >   Because whenever you save floating point images, it is 
converted into
>   >   unsigned char image.
>   >   So, the data will be lost.
>   > 
>   >   To save eigenfaces, try saving it as a text file, or 
something..
>   > 
>   >   Regards, :D
>   >   Paramesh.
>   > 
>   >   --- In OpenCV@xxxxxxxxxxxxxxx, "yeewan84" <yeewan84@> wrote:
>   >   >
>   >   > Hi Paramesh,
>   >   > 
>   >   > Thanks for checking the code for me.
>   >   > 
>   >   > I've tried to calculate the weights without saving and 
loading.
>   >   > It can recognize though result not so good. But does this 
mean that 
>   >   > everytime I want to recognize an input image, I've to 
recalculate 
>   >   > the eigenfaces? I'll be slow.
>   >   > 
>   >   > I've tried with OpenCV inbuilt function. It works but the 
same 
>   >   > problem occur. I cant save n load the eigenfaces to 
calculate the 
>   >   > weight.
>   >   > 
>   >   > Is there anyway, I can save the eigenfaces? 
>   >   > 
>   >   > rgds,
>   >   > evon
>   >   > 
>   >   > --- In OpenCV@xxxxxxxxxxxxxxx, "Parameswaran" 
<cegparamesh@> 
>   >   > wrote:
>   >   > >
>   >   > > Sorry for being late...
>   >   > > 
>   >   > > Looking at your code, i dont find any problem in the 
algorithm
>   you 
>   >   > use.
>   >   > > Did you get your eigenfaces right?
>   >   > > Did it look like an eigenface?
>   >   > > 
>   >   > > Why dont you try once, to calculate weights without 
saving and 
>   >   > loading
>   >   > > the eigenface?
>   >   > > 
>   >   > > And did you take a look at the inbuilt functions for 
doing PCA 
>   >   > using
>   >   > > OpenCV?
>   >   > > Using CalcEigenObjects and CalcDecompCoeff?
>   >   > > If you didnt try that one, download the PCA.zip 
attachment in the
>   >   > > group's files section..
>   >   > > 
>   >   > > Regards, :D
>   >   > > Paramesh.
>   >   > > 
>   >   > > --- In OpenCV@xxxxxxxxxxxxxxx, "yeewan84" <yeewan84@> 
wrote:
>   >   > > >
>   >   > > > Hi Paramesh,
>   >   > > > 
>   >   > > > Does it have to do with how I save the eigenfaces that 
effect 
>   >   > the 
>   >   > > > weights calculation? After i calculate the eigenfaces, 
I use it 
>   >   > to 
>   >   > > > calculate the weights of the training set, then save 
the 
>   >   > eigenfaces. 
>   >   > > > 
>   >   > > > For recognition, i load the eigenfaces then calculate 
the
>   weight 
>   >   > of 
>   >   > > > input image. Izit b'cos the eigenfaces values was 
altered
>   when i 
>   >   > > > save it?
>   >   > > > 
>   >   > > > 
>   >   > > > rgds,
>   >   > > > evon
>   >   > > > --- In OpenCV@xxxxxxxxxxxxxxx, "yeewan84" <yeewan84@> 
wrote:
>   >   > > > >
>   >   > > > > Hi,
>   >   > > > > 
>   >   > > > > Izit because i put all image data to array?
>   >   > > > > 
>   >   > > > > Training:
>   >   > > > > 
>   >   > > > > //****load training set
>   >   > > > > float** inputImg;
>   >   > > > > inputImg = (float**) calloc(nObjects, sizeof
(float*));
>   >   > > > > 
>   >   > > > > while( image ){
>   >   > > > >   filename = imgPath + image;
>   >   > > > >   temp = cvLoadImage(filename.c_str(), 0);
>   >   > > > >   inputImg[count] = (float*) calloc(150*150, sizeof
(float));
>   >   > > > > 
>   >   > > > >   for(int y = 0; y < 150; y++)
{                        
>   >   > > >       
>   >   > > > >     for(int x = 0; x < 150; x++){
>   >   > > > >       if( y == 0 || y == 1){
>   >   > > > >         inputImg[count][(y*149)+x] =
>   ((uchar*)(temp->imageData 
>   >   > + 
>   >   > > > > temp->widthStep * y))[x];
>   >   > > > >       
>   >   > > > >         //add up all value for calculate average 
face later
>   >   > > > >         average[(y*149)+x] += inputImg[count][(y*149)
+x];
>   >   > > > > 
>   >   > > > >       }else{
>   >   > > > >         inputImg[count][((y-1)*149)+x] = ((uchar*)
(temp-
>   >   > >imageData 
>   >   > > > + 
>   >   > > > > temp->widthStep * y))[x];
>   >   > > > > 
>   >   > > > >         //add up all value for calculate average 
face later
>   >   > > > >         average[((y-1)*149)+x] += inputImg[count]
[((y-1)*149)
>   >   > +x];
>   >   > > > >       }
>   >   > > > >     }
>   >   > > > >   }
>   >   > > > >   count++;
>   >   > > > >   image = strtok(NULL, ";");
>   >   > > > > 
>   >   > > > > }
>   >   > > > > 
>   >   > > > > 
>   >   > > > > //******calculate average face
>   >   > > > > for(int y = 0; y < 150; y++){
>   >   > > > >   for(int x = 0; x < 150; x++){
>   >   > > > >     if( y == 0 || y == 1){
>   >   > > > >       average[(y*149)+x] /= nObjects;
>   >   > > > >       ((uchar*)(temp->imageData + temp->widthStep * 
y))[x] = 
>   >   > avg
>   >   > > > > [(y*149)+x];
>   >   > > > > 
>   >   > > > >     }else{
>   >   > > > >       average[((y-1)*149)+x] /= nObjects;
>   >   > > > >       ((uchar*)(temp->imageData + temp->widthStep * 
y))[x] = 
>   >   > avg
>   >   > > > [((y-
>   >   > > > > 1)*149)+x]; 
>   >   > > > >     }
>   >   > > > >   }
>   >   > > > > }
>   >   > > > > 
>   >   > > > > //*****calculate difference face
>   >   > > > > for(y = 0; y < nObjects; y++)
>   >   > > > >   for(x = 0; x < 150*150; x++)
>   >   > > > >    inputImg[y][x] -= average[x];
>   >   > > > > 
>   >   > > > > 
>   >   > > > > 
>   >   > > > > //****calculate covariance matrix
>   >   > > > > float** L;
>   >   > > > > L = (float**) calloc(nObjects, sizeof(float*));
>   >   > > > > for(int y = 0; y < nObjects; y++)
>   >   > > > > L[y] = (float*) calloc(nObjects, sizeof(float));
>   >   > > > > 
>   >   > > > > for(int y = 0; y < nObjects; y++)
>   >   > > > >   for(int x = 0; x < nObjects; x++)
>   >   > > > >     L[y][x] = calcDotProduct(inputImg[y], inputImg
[x]);
>   >   > > > > 
>   >   > > > >       
>   >   > > > > 
>   >   > > > > //***for eigenvector and eigenvalues, i use a math 
library to 
>   >   > > > compute
>   >   > > > > 
>   >   > > > > //normalize eigenvector ||u||
>   >   > > > > double distance = 0, eigen = 0;
>   >   > > > > for(int i = 0; i < 150*150; i++)
>   >   > > > >   distance += pow(eigenImg[i], 2);
>   >   > > > >   distance = sqrt(distance);
>   >   > > > >       
>   >   > > > > if(distance != 1)
>   >   > > > >   for( int i = 0; i < 150*150; i++)
>   >   > > > >     eigenImg[i] /= distance;
>   >   > > > > //the eigenface is scale by 10000 (*10000) before 
saving
>   >   > > > > 
>   >   > > > > 
>   >   > > > > //****calculate weight
>   >   > > > > float **weight;
>   >   > > > > weight = (float **)calloc (nObjects, sizeof(float 
*));
>   >   > > > >       
>   >   > > > > for(int i = 0; i < nObjects; i++){
>   >   > > > >   weight[i] = (float *)calloc (M, sizeof(float));
>   >   > > > >   for(int j=0; j < M; j++)
>   >   > > > >      weight[i][j] = calcDotProduct(eigenImg[j], 
inputImg[i]);
>   >   > > > > }
>   >   > > > > 
>   >   > > > > Recognition:
>   >   > > > > //load eigenfaces the same way as loading training 
set above
>   >   > > > > //then scale the eigenface (divide by 10000)
>   >   > > > > 
>   >   > > > > //calculate weight
>   >   > > > > for(int j=0; j < M; j++)
>   >   > > > >  weight[j] = calcDotProduct( eigenImg[j], 
test_image );
>   >   > > > > }
>   >   > > > > 
>   >   > > > > 
>   >   > > > > //euclidean distance
>   >   > > > > for(int j=0; j < nObjects; j++)
>   >   > > > >  for(int k=0; k < M; j++)
>   >   > > > >   x1 += pow((weight[k] - weight_in_db[j][k], 2);
>   >   > > > > 
>   >   > > > > distance = sqrt(x1);
>   >   > > > > 
>   >   > > > > 
>   >   > > > > Thanks :)
>   >   > > > > 
>   >   > > > > rgds,
>   >   > > > > evon
>   >   > > > > 
>   >   > > > > --- In OpenCV@xxxxxxxxxxxxxxx, "Parameswaran" 
<cegparamesh@> 
>   >   > > > > wrote:
>   >   > > > > >
>   >   > > > > > Your weights and euclidean distance seems to be 
correct.
>   >   > > > > > The Euclidean distance is not too big...
>   >   > > > > > 
>   >   > > > > > >I dont get zero for euclidean distance when I use 
the same 
>   >   > input
>   >   > > > > > >image for calculating eigenface.
>   >   > > > > > >
>   >   > > > > > >Whatever test image I put to recognize, I always 
get back 
>   >   > that 
>   >   > > > few
>   >   > > > > > >face for nearest face. For example, I have face 
class 1 to 
>   >   > 10, 
>   >   > > > the
>   >   > > > > > >system always give me face class 2 and 5 for 
nearest face 
>   >   > > > classes.
>   >   > > > > > 
>   >   > > > > > This is where i doubt something.
>   >   > > > > > The Euclidean distance should be zero if the test 
image is 
>   >   > the 
>   >   > > > > same as
>   >   > > > > > the face database.
>   >   > > > > > 
>   >   > > > > > Well. What can i suggest you now?
>   >   > > > > > Take a look at your code in detail.
>   >   > > > > > If you are calculating the weights and euclidean 
distance 
>   >   > the 
>   >   > > > same 
>   >   > > > > way
>   >   > > > > > for the test image, then Euclidean distance should 
be zero.
>   >   > > > > > What is the euclidean distance when you get, if 
you give
>   the 
>   >   > test
>   >   > > > > > image as the same on in database?
>   >   > > > > > Is it less than 1000 or 2000? It should be..
>   >   > > > > > Else something went wrong somewhere.
>   >   > > > > > 
>   >   > > > > > Regards, :D
>   >   > > > > > Paramesh.
>   >   > > > > > 
>   >   > > > > > PS: Last but not the least, if you cant find the 
error, 
>   >   > paste the
>   >   > > > > > entire code. I'll look through it and try to find 
the error 
>   >   > if i 
>   >   > > > > can.
>   >   > > > > > 
>   >   > > > > > 
>   >   > > > > > 
>   >   > > > > > --- In OpenCV@xxxxxxxxxxxxxxx, "yeewan84" 
<yeewan84@> wrote:
>   >   > > > > > >
>   >   > > > > > > Hi,
>   >   > > > > > > 
>   >   > > > > > > Sample weights:
>   >   > > > > > > -112.83, -1067.05, 1654.81, -4087.68, 2575.40, -
1806.17, 
>   >   > > > 1013.89
>   >   > > > > > >  
>   >   > > > > > > Sample euclidean distance:
>   >   > > > > > > 8599.753906, 9766.369141, 9910.776367, 
9836.765625, 
>   >   > > > 11140.266602,
>   >   > > > > > > 10934.599609, 9329.093750
>   >   > > > > > > (izit too big?)
>   >   > > > > > > 
>   >   > > > > > > this is how i calculate the distance:
>   >   > > > > > > for(int i = 0; i < M; i++){
>   >   > > > > > >    x += pow((input_weight[k] - weight[k]), 2);
>   >   > > > > > > }
>   >   > > > > > > distance = sqrt(x);
>   >   > > > > > > 
>   >   > > > > > > Sample eigenfaces image data:
>   >   > > > > > > 6.44592935e-010, 2.57854227e-009, 4.14895069e-
008,
>   >   > > > > > > 1.03723776e-008, 9.90167757e-012, 2.59315036e-
009,
>   >   > > > > > > 1.59329189e-010, 1.66892988e-007, 6.78776985e-
007,
>   >   > > > > > > (I've normalize the eigenfaces to be in unit 
length, 
>   >   > ||u|| )
>   >   > > > > > > 
>   >   > > > > > > Thank you very much :)
>   >   > > > > > > 
>   >   > > > > > > rgds,
>   >   > > > > > > evon
>   >   > > > > > >
>   >   > > > > >
>   >   > > > >
>   >   > > >
>   >   > >
>   >   >
>   > 
>   > 
>   > 
>   > 
>   > 
>   > 
>   >   Change settings: http://www.yahoogroups.com/mygroups, select
>   >      Get Emails (get all posts)
>   >      Daily Digest (one summary email per day)
>   >      Read on the web (read posts on the web only)Or 
Unsubscribe by
>   mailing OpenCV-unsubscribe@xxxxxxxxxxxxxxx
>   > 
>   > 
>   > 
>   > 
>   >   SPONSORED LINKS Open source software  Computer vision  
>   > 
>   > 
>   >
>   -----------------------------------------------------------------
-------------
>   >   YAHOO! GROUPS LINKS 
>   > 
>   >     a..  Visit your group "OpenCV" on the web.
>   >       
>   >     b..  To unsubscribe from this group, send an email to:
>   >      OpenCV-unsubscribe@xxxxxxxxxxxxxxx
>   >       
>   >     c..  Your use of Yahoo! Groups is subject to the Yahoo! 
Terms of
>   Service. 
>   > 
>   > 
>   >
>   -----------------------------------------------------------------
-------------
>   > 
>   > 
>   > 
>   > [Non-text portions of this message have been removed]
>   >
> 
> 
> 
> 
> 
> 
> 
>   Change settings: http://www.yahoogroups.com/mygroups, select
>      Get Emails (get all posts)
>      Daily Digest (one summary email per day)
>      Read on the web (read posts on the web only)Or Unsubscribe by 
mailing OpenCV-unsubscribe@xxxxxxxxxxxxxxx
> 
> 
> 
> 
> -------------------------------------------------------------------
-----------
>   YAHOO! GROUPS LINKS 
> 
>     a..  Visit your group "OpenCV" on the web.
>       
>     b..  To unsubscribe from this group, send an email to:
>      OpenCV-unsubscribe@xxxxxxxxxxxxxxx
>       
>     c..  Your use of Yahoo! Groups is subject to the Yahoo! Terms 
of Service. 
> 
> 
> -------------------------------------------------------------------
-----------
> 
> 
> 
> [Non-text portions of this message have been removed]
>






Change settings: http://www.yahoogroups.com/mygroups, select
   Get Emails (get all posts)
   Daily Digest (one summary email per day)
   Read on the web (read posts on the web only)Or Unsubscribe by mailing 
OpenCV-unsubscribe@xxxxxxxxxxxxxxx