100 likes | 279 Views
Adam Yeh. UCF Computer Vision REU Week 7. Previous. Finished expression vs. expression classification Attempted one vs. all classification (using libsvm implementation) Failed to classify anything Started work on new algorithm. Binary SVM results. Algorithm:
E N D
Adam Yeh UCF Computer Vision REU Week 7
Previous • Finished expression vs. expression classification • Attempted one vs. all classification (using libsvm implementation) • Failed to classify anything • Started work on new algorithm
Algorithm: • Run SVM for every pair of classes • Record all the non-support vectors from each binary SVM training • For each class, find the intersection of NSVs from all the binary training • For each test image, which group of NSVs it is “closest” to; the image is then classified into that group • Use centroid of each NSV group to represent group • Calculate Euclidian distance
Progress • Code so far: • Trains all binary SVMs automatically • Isolates each set of NSVs from the binary SVMs • Finds intersection of the NSVs • Next: • Method to determine which class a test image is closest to • Centroid/euclidian distance • Mahalanobis distance
Next week • Finish coding new algorithm • Start testing on facial expressions