SAIVT Thermal Feature Detection


The SAIVT-Thermal Feature Detection Database contains a number of images suitable for evaluating the performance of feature detection and matching in the thermal image domain.

The database includes conditions unique to the thermal domain such as non-uniformity noise; as well as condition common to other domains such as viewpoint changes, and compression and blur.

You can read our paper on eprints.

Contact Dr Simon Denman for further information.


The SAIVT Thermal Feature Detection Database is © 2012 QUT and is licensed under the Creative Commons Attribution-ShareAlike 3.0 Australia License.


To attribute this database, please include the following citation:

An exploration of feature detector performance in the thermal-infrared modality. Vidas, Stephen, Lakemond, Ruan, Denman, Simon, Fookes, Clinton B., Sridharan, Sridha, & Wark, Tim. (2011) In Bradley, Andrew, Jackway, Paul, Gal, Yaniv, & Salvado, Olivier (Eds.) Proceedings of the 2011 International Conference on Digital Image Computing: Techniques and Applications, IEEE , Sheraton Noosa Resort & Spa, Noosa, QLD, pp. 217-223.

Acknowledging the database in your publications

In addition to citing our paper, we kindly request that the following text be included in an acknowledgements section at the end of your publications:

We would like to thank the SAIVT Research Labs at Queensland University of Technology (QUT) for freely supplying us with the SAIVT Thermal Feature Detection Database for our research.

Installing the database

Download and unzip the following archive:

 SAIVT-ThermalFeatureDetection.tar.gz (187MB)

A copy of the publication can be found at, and is also included in this package (Vidas 2011 – An exploration of feature detector performance in the thermal-infrared modality.pdf).

Related publications of interest may be found on the following webpages:

The database has the following structure:

  • Each of the ten environments is allocated its own directory.
  • Within most of these directories, thermal-infrared and visible-spectrum data is separated into the “thermal” and “visible” subdirectories respectively
  • Within each of these subdirectories, a “profile” folder is present which contains a sequence of “ideal” (untransformed) images in 8-bit depth format.
  • The “thermal” subdirectories also contain a “pure” folder which contains identical images in their original 16-bit depth format (which is difficult to visualize).
  • Also within each “thermal” subdirectory there may be additional folders present.

Each of these folders contain images under a single, controlled image transformation, the acronyms for which are expanded at the end of this document. The level of transformation varies (generally increasing in severity) as the numerical label for each subfolder increases.


  • CMP Image Compression
  • GAU Gaussian Noise
  • NRM Histogram Normalization
  • NUC Non-Uniformities Noise
  • OFB Out-of-focus Blur
  • QNT Quantization Noise
  • ROT Image Rotation
  • SAP Salt and Pepper Noise
  • TOD Time of day variation
  • VPT Viewpoint change