Incorporating SIFT with hard C-means algorithm
The scale invariant feature transform (SIFT) has been used widely as a tool in object recognition. However, when there are several keyframes for one object in the training database, the number of keypoint descriptors for that object might be huge. The matching process of a test keypoint has to be do...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Conference or Workshop Item |
| Language: | English |
| Published: |
2014
|
| Online Access: | http://www.scopus.com/inward/record.url?eid=2-s2.0-77952635713&partnerID=40&md5=441af0192450f31f365fb99c5d5cc2db http://cmuir.cmu.ac.th/handle/6653943832/1513 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Chiang Mai University |
| Language: | English |
| Summary: | The scale invariant feature transform (SIFT) has been used widely as a tool in object recognition. However, when there are several keyframes for one object in the training database, the number of keypoint descriptors for that object might be huge. The matching process of a test keypoint has to be done on all keypoints in the training database, hence, the amount of matching time is huge. Since the keyframes in the training database are from the same object, there must be some keypoints that are similar. In this paper we incorporate SIFT with the Hard C-Means (HCM) algorithm to group keypoint descriptors and then utilize the prototypes in the matching process instead. We implement this algorithm with three data sets, i.e., bottle, MPEG 7 and Thai hand gesture. We found that on the bottle and MPEG 7 test data sets, the algorithm outperform the one with SIFT with much smaller matching computation time. For the Thai hand gesture data set, the correct classification with much less matching times on the test data set from the proposed algorithm is comparable with that of SIFT. ©2010 IEEE. |
|---|