Presentation | 2012-11-17 A model of view prediction in three dimensional object recognition Toru UENO, Nobuhiko ASAKURA, Takafumi SASAOKA, Toshio INUI, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Previous studies have identified two processes for view-invariant 3-D object recognition. One is a view-based process that uses the 2-D image features of an object from a particular viewpoint, and the other is a structural-description process that includes verification of the spatial relations between distinct parts of the object. The view-based process requires generalization from stored view information to novel views. In this study, we model the mechanism of view generalization as a form of view prediction, and propose a neural network based on the generalized radial basis function (GRBF) network. The input of the network is a pair of 2-D views of a paper-clip object. These views are separated by a 10 degree rotation along the vertical axis. We trained the network to output a view of the same object after a further 10 degree rotation. Following training, the network was capable of predicting rotated views not only for a set of training objects but also for untrained objects. We then introduced a feedback loop from the output to the input, in order to investigate whether the network could predict views with larger rotations. The prediction error increased with the amount of rotation, but a prediction accuracy of about 80% was still achieved with rotations of up to 30 degrees. We demonstrated that the accuracy as a function of rotations is consistent with the human performance in the recognition experiment using paper-clip objects reported in the previous study. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Object recognition / Mental rotation / GRBF neural network |
Paper # | NC2012-70 |
Date of Issue |
Conference Information | |
Committee | NC |
---|---|
Conference Date | 2012/11/9(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | |
Vice Chair | |
Secretary | |
Assistant |
Paper Information | |
Registration To | Neurocomputing (NC) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A model of view prediction in three dimensional object recognition |
Sub Title (in English) | |
Keyword(1) | Object recognition |
Keyword(2) | Mental rotation |
Keyword(3) | GRBF neural network |
1st Author's Name | Toru UENO |
1st Author's Affiliation | Graduate School of Informatics, Kyoto University() |
2nd Author's Name | Nobuhiko ASAKURA |
2nd Author's Affiliation | Graduate School of Informatics, Kyoto University |
3rd Author's Name | Takafumi SASAOKA |
3rd Author's Affiliation | Graduate School of Informatics, Kyoto University |
4th Author's Name | Toshio INUI |
4th Author's Affiliation | Graduate School of Informatics, Kyoto University |
Date | 2012-11-17 |
Paper # | NC2012-70 |
Volume (vol) | vol.112 |
Number (no) | 298 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |