Файл:Deconvolutional networks Cvpr2010.pdf

Matthew D. Zeiler, Dilip Krishnan, Graham W. Taylor and Rob Fergus Dept. of Computer Science, Courant Institute, New York University {zeiler,dilip,gwtaylor,fergus}@cs.nyu.edu

Abstract

Building robust low and mid-level image representations, beyond edge primitives, is a long-standing goal in vision. Many existing feature detectors spatially pool edge informationwhichdestroyscuessuchasedgeintersections, parallelism and symmetry. We present a learning frameworkwherefeaturesthatcapturethesemid-levelcuesspontaneously emerge from image data. Our approach is based ontheconvolutionaldecompositionofimagesunderasparsity constraint and is totally unsupervised. By building a hierarchy of such decompositions we can learn rich feature setsthatarearobustimagerepresentationforboththeanalysis and synthesis of images.