DocumentCode :
2916347
Title :
High-dimensional signature compression for large-scale image classification
Author :
Sánchez, Jorge ; Perronnin, Florent
fYear :
2011
fDate :
20-25 June 2011
Firstpage :
1665
Lastpage :
1672
Abstract :
We address image classification on a large-scale, i.e. when a large number of images and classes are involved. First, we study classification accuracy as a function of the image signature dimensionality and the training set size. We show experimentally that the larger the training set, the higher the impact of the dimensionality on the accuracy. In other words, high-dimensional signatures are important to obtain state-of-the-art results on large datasets. Second, we tackle the problem of data compression on very large signatures (on the order of 105 dimensions) using two lossy compression strategies: a dimensionality reduction technique known as the hash kernel and an encoding technique based on product quantizers. We explain how the gain in storage can be traded against a loss in accuracy and/or an increase in CPU cost. We report results on two large databases - ImageNet and a dataset of lM Flickr images - showing that we can reduce the storage of our signatures by a factor 64 to 128 with little loss in accuracy. Integrating the decompression in the classifier learning yields an efficient and scalable training algorithm. On ILSVRC2010 we report a 74.3% accuracy at top-5, which corresponds to a 2.5% absolute improvement with respect to the state-of-the-art. On a subset of 10K classes of ImageNet we report a top-1 accuracy of 16.7%, a relative improvement of 160% with respect to the state-of-the-art.
Keywords :
handwriting recognition; image classification; image coding; learning (artificial intelligence); vector quantisation; classifier learning; data compression; dimensionality reduction technique; encoding technique; hash kernel technique; image classification; lM Flickr images; lossy compression strategy; product quantizers; scalable training algorithm; signature compression; Accuracy; Data compression; Encoding; Image coding; Kernel; Training; Vectors;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Computer Vision and Pattern Recognition (CVPR), 2011 IEEE Conference on
Conference_Location :
Providence, RI
ISSN :
1063-6919
Print_ISBN :
978-1-4577-0394-2
Type :
conf
DOI :
10.1109/CVPR.2011.5995504
Filename :
5995504
Link To Document :
بازگشت