Multiclass Object Recognition with Sparse, Localized Features

@inproceedings{citeulike:965068, abstract = {We apply a biologically inspired model of visual object recognition to the multiclass object categorization problem. Our model modifies that of Serre, Wolf, and Poggio. As in that work, we first apply Gabor filters at all positions and scales; feature complexity and position/scale invariance are then built up by alternating template matching and max pooling operations. We refine the approach in several biologically plausible ways, using simple versions of sparsification and lateral inhibition. We demonstrate the value of retaining some position and scale information above the intermediate feature level. Using feature selection we arrive at a model that performs better with fewer features. Our final model is tested on the Caltech 101 object categories and the UIUC car localization task, in both cases achieving state-of-the-art performance. The results strengthen the case for using this class of model in computer vision.}, author = {Mutch, J. and Lowe, D. G. }, citeulike-article-id = {965068}, journal = {Computer Vision and Pattern Recognition, 2006 IEEE Computer Society Conference on}, keywords = {bag-of-features, object-categorization, object-recognition, photo2search, reading, spatial-relationship}, pages = {11--18}, posted-at = {2007-03-09 11:30:54}, priority = {5}, title = {Multiclass Object Recognition with Sparse, Localized Features}, url = {http://ieeexplore.ieee.org/xpls/abs\_all.jsp?arnumber=1640736}, volume = {1}, year = {2006} }

TY - CONF ID - citeulike:965068 L3 - citeulike-article-id:965068 TI - Multiclass Object Recognition with Sparse, Localized Features JF - Computer Vision and Pattern Recognition, 2006 IEEE Computer Society Conference on VL - 1 SP - 11 EP - 18 N2 - We apply a biologically inspired model of visual object recognition to the multiclass object categorization problem. Our model modifies that of Serre, Wolf, and Poggio. As in that work, we first apply Gabor filters at all positions and scales; feature complexity and position/scale invariance are then built up by alternating template matching and max pooling operations. We refine the approach in several biologically plausible ways, using simple versions of sparsification and lateral inhibition. We demonstrate the value of retaining some position and scale information above the intermediate feature level. Using feature selection we arrive at a model that performs better with fewer features. Our final model is tested on the Caltech 101 object categories and the UIUC car localization task, in both cases achieving state-of-the-art performance. The results strengthen the case for using this class of model in computer vision. KW - bag-of-features KW - object-categorization KW - object-recognition KW - photo2search KW - reading KW - spatial-relationship AU - Mutch, J AU - Lowe, DG PY - 2006/// UR - http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1640736 ER -