Overview
The database was created as a part of the paper "An Evaluation of Local Feature Detectors and Descriptors for Infrared Images" as a part of ECCV Workshop on LOCAL FEATURES: STATE OF THE ART, OPEN PROBLEMS AND PERFORMANCE EVALUATION (2016).
The database, as of publishing the paper, consist of 12 different scenes and 6 different geometric and photometric transformations and a total of 118 images. The six transformations are:
- View point (Perspective changes)
- Scale (Zoom)
- Rotation
- Blur
- Noise
- Downsampling
For each scene a reference image and its test images are related by a 2D-linear transformation, a homography. Usage of homographies hold if the content of the image is a planar scene or for camera rotations around its center. By applying the corresponding homography to a test image it is transformed to the same situation as the reference image.
The database is publicly available in the compressed IRImagesDataset.zip file under Downloads. All images (.ppm) as well as homographies (.mat) are included. The exception for homographies are for the transformations blur and noise where the relationship is an identity 3x3 dimensional matrix and is therefore not included.
Papers
The dataset was introduced in the the following paper:
An Evaluation of Local Feature Detectors and Descriptors for Infrared Images
Johan Johansson, Martin Solli and Atsuto Maki
FLIR Systems AB, KTH
ECCV Workshop on LOCAL FEATURES: STATE OF THE ART, OPEN PROBLEMS AND PERFORMANCE EVALUATION (2016)
[paper, 2.8 MB]