image processing - Hand Posture Recognition Using SURF with Adaptive Boosting -
i´m trying implement algorithm of paper:
http://www.bmva.org/bmvc/2012/ws/paper5.pdf
it´s "training process target postures" algorithm, in page 6.
basicly technique use adaboost , surf (extractor of features) hand posture recognition. wrote before, i´m trying implement still didn´t understand algorithm. problem question have, like:
- what matching score list, mention in page 5, last paragraph?
- what relationship between threshold , training process (trainning algorithm)?
- in line 7 of training algorithm: return of function ht(ix,ft,tet)?
i know if have algorithm implemented or if can me in questions.
thank answer (help), because i´m investing time in algorithm , still didn´t understand algorithm implement.
scanning through paper, here inferred. please note did not go through paper in detail might wrong, having implement adaboost before, can clarify doubts:
- during testing part, assign matching 'score' each posture in database. want determine posture of test image accurately possible. 'score' represents how 'close' test posture posture in database. posture highest score in end result of matching/testing.
2 , 3. during selecting weak classifiers, have threshold iteration selecting weak classifiers stops weight becomes less threshold. threshold should optimum in 'selected' weak classifier should unique enough able classify corresponding vector efficiently. @ same time, should not strict on threshold or time taken iterations large. hence, threshold should optimum, , paper shows how calculate it. answers second question. function ht talking weak classifier.
Comments
Post a Comment