Sparse-Coded Features for Image Retrieval
published: April 3, 2014, recorded: September 2013, views: 2913
Report a problem or upload filesIf you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.
State-of-the-art image retrieval systems typically represent an image with a bag of
low-level features. Since different images often exhibit different kinds of low-level characteristics,
it is desirable to represent an image with multiple types of complementary
features. The systems scalability is, however, significantly lowered when increasing the
number of feature types, as the amount of data is also increased rapidly both in index and
in query representation.
In this paper, we apply sparse coding to derive a compact yet discriminative image representation from multiple types of features for large-scale image retrieval. We first convert each feature descriptor into a sparse code, and aggregate each type of sparsecoded features into a single vector by max-pooling. Multiple vectors from different types of features are then concatenated and compressed to obtain the final representation. Our approach allows us to add more types of features to improve discriminability without sacrificing scalability. In particular, we design a new micro feature which is complementary to existing local invariant features. By combining our micro feature with various local invariant features using the sparse-coding framework, our final compact representation outperforms the state of the art both in retrieval performance and in scalability.
Link this pageWould you like to put a link to this lecture on your homepage?
Go ahead! Copy the HTML snippet !