Share Email Print

Proceedings Paper

A novel class sensitive hashing technique for large-scale content-based remote sensing image retrieval
Author(s): Thomas Reato; Begüm Demir; Lorenzo Bruzzone
Format Member Price Non-Member Price
PDF $17.00 $21.00

Paper Abstract

This paper presents a novel class sensitive hashing technique in the framework of large-scale content-based remote sensing (RS) image retrieval. The proposed technique aims at representing each image with multi-hash codes, each of which corresponds to a primitive (i.e., land cover class) present in the image. To this end, the proposed method consists of a three-steps algorithm. The first step is devoted to characterize each image by primitive class descriptors. These descriptors are obtained through a supervised approach, which initially extracts the image regions and their descriptors that are then associated with primitives present in the images. This step requires a set of annotated training regions to define primitive classes. A correspondence between the regions of an image and the primitive classes is built based on the probability of each primitive class to be present at each region. All the regions belonging to the specific primitive class with a probability higher than a given threshold are highly representative of that class. Thus, the average value of the descriptors of these regions is used to characterize that primitive. In the second step, the descriptors of primitive classes are transformed into multi-hash codes to represent each image. This is achieved by adapting the kernel-based supervised locality sensitive hashing method to multi-code hashing problems. The first two steps of the proposed technique, unlike the standard hashing methods, allow one to represent each image by a set of primitive class sensitive descriptors and their hash codes. Then, in the last step, the images in the archive that are very similar to a query image are retrieved based on a multi-hash-code-matching scheme. Experimental results obtained on an archive of aerial images confirm the effectiveness of the proposed technique in terms of retrieval accuracy when compared to the standard hashing methods.

Paper Details

Date Published: 4 October 2017
PDF: 9 pages
Proc. SPIE 10427, Image and Signal Processing for Remote Sensing XXIII, 1042712 (4 October 2017); doi: 10.1117/12.2279407
Show Author Affiliations
Thomas Reato, Univ. degli Studi di Trento (Italy)
Begüm Demir, Univ. degli Studi di Trento (Italy)
Lorenzo Bruzzone, Univ. degli Studi di Trento (Italy)

Published in SPIE Proceedings Vol. 10427:
Image and Signal Processing for Remote Sensing XXIII
Lorenzo Bruzzone, Editor(s)

© SPIE. Terms of Use
Back to Top
Sign in to read the full article
Create a free SPIE account to get access to
premium articles and original research
Forgot your username?