Name
Abstract | ||
|---|---|---|
Hough Forest is an object detection method based on voting from patch images. In the Hough Forest training, some negative patches are trained as a positive sample because the patches are truncated from the background region in a positive image. This makes a reason to occur false positives. To overcome this problem, we introduce weight updating of training sample to the Hough Forest. In the training of the proposed method, if there is a positive sample with a high value of similarity with a negative sample, sample weight are updated to be smaller at each layer of the decision tree. This makes it possible to suppress the vote to the background area. Experimental results show that the detection performance of the proposed method is 11% better than that of conventional method, and is same as the conventional method with masked images. |
Year | DOI | Venue |
|---|---|---|
2015 | 10.1109/MVA.2015.7153148 | Machine Vision Applications |
Keywords | Field | DocType |
feature extraction,object detection,Hough Forest training,negative patches,object detection method,weighted Hough Forest | Training set,Decision tree,Object detection,Computer vision,Pattern recognition,Voting,Computer science,Sample Weight,Hough transform,Feature extraction,Artificial intelligence,False positive paradox | Conference |
Citations | PageRank | References |
1 | 0.36 | 6 |
Authors | ||
4 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Yosuke Murai | 1 | 1 | 0.36 |
| Yuji Yamauchi | 2 | 43 | 10.45 |
| Takayoshi Yamashita | 3 | 377 | 46.83 |
| fujiyoshi | 4 | 730 | 101.43 |