Why do linear SVMs trained on HOG features perform so well?

Linear Support Vector Machines trained on HOG features are now a de facto standard across many visual perception tasks. Their popularisation can largely be attributed to the step-change in performance they brought to pedestrian detection, and their subsequent successes in deformable parts models. This paper explores the interactions that make the HOG-SVM symbiosis perform so well. By connecting the feature extraction and learning processes rather than treating them as disparate plugins, we show that HOG features can be viewed as doing two things: (i) inducing capacity in, and (ii) adding prior to a linear SVM trained on pixels. From this perspective, preserving second-order statistics and locality of interactions are key to good performance. We demonstrate surprising accuracy on expression recognition and pedestrian detection tasks, by assuming only the importance of preserving such local second-order interactions.

[1]  Geoffrey E. Hinton.  Learning multiple layers of representation , 2007, Trends in Cognitive Sciences.

[2]  Honglak Lee,et al.  Convolutional deep belief networks for scalable unsupervised learning of hierarchical representations , 2009, ICML '09.

[3]  Yongsheng Gao,et al.  Face recognition across pose: A review , 2009, Pattern Recognit..

[4]  J. Daugman.  Uncertainty relation for resolution in space, spatial frequency, and orientation optimized by two-dimensional visual cortical filters. , 1985, Journal of the Optical Society of America. A, Optics and image science.

[5]  Eero P. Simoncelli,et al.  Natural image statistics and neural representation. , 2001, Annual review of neuroscience.

[6]  Oscar Déniz-Suárez,et al.  Face recognition using Histograms of Oriented Gradients , 2011, Pattern Recognit. Lett..

[7]  Jitendra Malik,et al.  Discriminative Decorrelation for Clustering and Classification , 2012, ECCV.

[8]  Charless C. Fowlkes,et al.  Do We Need More Training Data or Better Models for Object Detection? , 2012, BMVC.

[9]  Marc'Aurelio Ranzato,et al.  Learning invariant features through topographic filter maps , 2009, CVPR.

[10]  Simon Lucey,et al.  V1-Inspired Features Induce a Weighted Margin in SVMs , 2012, ECCV.

[11]  James J. DiCarlo,et al.  How Does the Brain Solve Visual Object Recognition? , 2012, Neuron.

[12]  Y-Lan Boureau,et al.  Learning Convolutional Feature Hierarchies for Visual Recognition , 2010, NIPS.

[13]  Stephen P. Boyd,et al.  Distributed Optimization and Statistical Learning via the Alternating Direction Method of Multipliers , 2011, Found. Trends Mach. Learn..

[14]  Fatih Murat Porikli,et al.  Region Covariance: A Fast Descriptor for Detection and Classification , 2006, ECCV.

[15]  Bill Triggs,et al.  Histograms of oriented gradients for human detection , 2005, 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05).

[16]  Yi Yang,et al.  Articulated pose estimation with flexible mixtures-of-parts , 2011, CVPR 2011.

[17]  R. Fergus,et al.  Learning invariant features through topographic filter maps , 2009, 2009 IEEE Conference on Computer Vision and Pattern Recognition.

[18]  David A. McAllester,et al.  Object Detection with Discriminatively Trained Part Based Models , 2010, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[19]  Krista A. Ehinger,et al.  SUN database: Large-scale scene recognition from abbey to zoo , 2010, 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.

[20]  Chih-Jen Lin,et al.  A dual coordinate descent method for large-scale linear SVM , 2008, ICML '08.

[21]  M. Carandini,et al.  Normalization as a canonical neural computation , 2011, Nature Reviews Neuroscience.

[22]  A. Hyvärinen,et al.  Complex cell pooling and the statistics of natural images , 2007, Network.