Randomized Clustering Forests for Image Classification
Frederic Jurie (INRIA Grenoble, France)
This talk will be focused on 3 recent contributions to the problems of
image classification and image search. Some of the most effective
recent methods for content-based image classification work by
extracting image descriptors, quantizing them according to a coding
rule such as k-means vector quantization, accumulating histograms of
the resulting visual word codes over the image, and classifying these
with a conventional classifier such as an SVM. Large numbers of
descriptors and large codebooks are required for good results and this
becomes slow using k-means. First, a new paradigm for representing
images will be presented. We will introduce Extremely Randomized
Clustering Forests -- ensembles of randomly created clustering trees
-- and show that they provide more accurate results, much faster
training and testing and good resistance to background clutter.
Second, an efficient image classification method will be described. It
combines ERC-Forests and saliency maps very closely with the
extraction of image information. It will be shown that this method,
which requires to extract 20 times less information from images than
the standard bag-of-word algorithm also provides more accurate
results. in several state-of-the-art image classification tasks.
Finally, we will show that the proposed ERC-Forests can also be used
very successfully for learning distance between images. The
similarity measure has been evaluated on four very different
datasets and always outperforms the state-of-the-art competitive
approaches.