Abstract

Dictionary learning algorithms or supervised deep convolution networks have considerably improved the effifi- ciency of predefifined feature representations such as SIFT. We introduce a deep scattering convolution network, with complex wavelet fifilters over spatial and angular variables. This representation brings an important improvement to results previously obtained with predefifined features over object image databases such as Caltech and CIFAR. The resulting accuracy is comparable to results obtained with unsupervised deep learning and dictionary based representations. This shows that refifining image representations by using geometric priors is a promising direction to improve image classifification and its understanding.