Hi ,
I have more than 4 type of things (knife, spoon, fork etc...) train images. I have to obtain different feature extraction algorithm performance and different learning algorithm (ie. SIFT, SURF, BRISK features to use SVM, MLP, etc.). When I use SIFT, SURF they return descriptors as matrix(features of keypoints ) for one type of train image. How can I down grade or select among features of this matrix to row vector (featurtes) that enough to train MLP or SVM ? For example SIFT algorithm return 100 keypoints and each keypoint have 128 feature for one type of things of train image. It means 100 rows x 128 column matrix. What is the next step ?
↧