Evaluation of feature point detection in high dynamic range imagery

• We study suitability of HDR and tone mapped imagery for detection of feature points.
• We test 6 detectors using 16 image formats under various scene changes.
• Results show that current FP detectors are not well tuned to process HDR images.
• Tone mapping is the best contemporary solution for FP detection in HDR imagery.
• A gradient-based local tone mapper with contrasts boosting produces best results.

This paper evaluates the suitability of High Dynamic Range (HDR) imaging techniques for Feature Point (FP) detection under demanding lighting conditions. The FPs are evaluated in HDR, tone mapped HDR, and traditional Low Dynamic Range (LDR) images. Eleven global and local tone mapping operators are evaluated and six widely used FP detectors are used in the experiments (Harris, Shi–Tomasi, DoG, Fast Hessian, FAST, and BRISK). The distribution and repeatability rate of FPs are studied under changes of camera viewpoint, camera distance, and scene lighting. The results of the experiments show that current FP detectors cannot cope with HDR images well. The best contemporary solution is thus tone mapping of HDR images using a local tone mapper as a pre-processing step.