This is a sample implementation for using svm-multiclass by cornell for image classification based on the test images. https://www.cs.cornell.edu/people/tj/svm_light/svm_multiclass.html
We intially resize all the images to fixed size so that we have feature vectors of same size for all the images.
##Train
Feature vectors of the image is written int the format as below in the file named train.dat
class 1:value 2:value 3:value ....
call the svm_multiclass as
system("./svm_multiclass_learn -c 1.0 train.dat model ");
1.0 is regularization parameter
This will generate a model file which wil be used in the test phase
##Test
Feature vectors of the image is written int the format as below in the file named test.dat
class 1:value 2:value 3:value ....
call the svm_multiclass as
system("./svm_multiclass_classify test.dat model predictions")
This will give the ouput predictions