Thread Subject: support vector machines

Hi,
I have been trying to classify indoor and outdoor images using SVM. But I'm kind of stuck in this.After training a set of images which has both indoor and outdoor images with their luminance n chrominance values, I happened to get 3 or 4 support vectors.But I dont kow how to proceed further..How am I supposed to test another set of images using the trained datasets or the obtained support vector values.
Can any1 help me plzz...

"rishtah " <rishtah@mathworks.com> wrote in message <h85mnr$4np$1@fred.mathworks.com>...
> Hi,
> I have been trying to classify indoor and outdoor images using SVM. But I'm kind of stuck in this.After training a set of images which has both indoor and outdoor images with their luminance n chrominance values, I happened to get 3 or 4 support vectors.But I dont kow how to proceed further..How am I supposed to test another set of images using the trained datasets or the obtained support vector values.
> Can any1 help me plzz...

What kind of code do you use? Is it linear, soft SVM, non-linear kernel-trick SVM? etc... Probably reading careful the documentation of your code you are using can answer right the way.

Bruno

Hi,

I've kind of figured little bit on training and testing data using svm,but still I'm missing something here. Basically,I'm using the linear kernal method. After training the data, I calculated my classperf value etc..But for the images that i tested, how can I know under which class they were classified in svm...is there a way to find out the predicted classes in matlab?? I read the documentation n tried finding on google..but cudnt find anything related to SVM...
thanks...

"rishtah " <rishtah@mathworks.com> wrote in message <h8a6qh$i7r$1@fred.mathworks.com>...
> Hi,
>
> I've kind of figured little bit on training and testing data using svm,but still I'm missing something here. Basically,I'm using the linear kernal method. After training the data, I calculated my classperf value etc..But for the images that i tested, how can I know under which class they were classified in svm...is there a way to find out the predicted classes in matlab?? I read the documentation n tried finding on google..but cudnt find anything related to SVM...
> thanks...

the vector w,b is computed as

w = sum_{on all data} lambda*y*x
b = -1/2 { mean_{on_SV suchthat y<0)} w'*x +
               mean_[on_SV_sucththat y>0) w'*x }

lambda is the dual variable returned after the training
SV are support vector
x is your trained data
y is the classification binary {-1,1}

The classification function is sign(w'*x + b)

There are a tone of papers that provide those formula.

Bruno