CP HDR: A feature point detection and description library for LDR and HDR images

TL;DR

The paper investigates feature point detection and description in HDR images, addressing limitations of LDR in extreme lighting. It introduces CP_HDR, a library that supports both LDR and HDR inputs and includes HDR-special variants HfHDR and SfHDR alongside SIFT/Harris baselines. Through a systematic review and controlled experiments, it shows HDR inputs improve detection uniformity and, for descriptors like SIFT, can boost mean average precision, with HDR-specific variants often performing best on UR and RR metrics. The work provides a practical, open-source toolkit for HDR-aware FP processing and highlights significant gains in FP distribution and matching performance, guiding future HDR-aware vision pipelines.

Abstract

In computer vision, characteristics refer to image regions with unique properties, such as corners, edges, textures, or areas with high contrast. These regions can be represented through feature points (FPs). FP detection and description are fundamental steps to many computer vision tasks. Most FP detection and description methods use low dynamic range (LDR) images, sufficient for most applications involving digital images. However, LDR images may have saturated pixels in scenes with extreme light conditions, which degrade FP detection. On the other hand, high dynamic range (HDR) images usually present a greater dynamic range but FP detection algorithms do not take advantage of all the information in such images. In this study, we present a systematic review of image detection and description algorithms that use HDR images as input. We developed a library called CP_HDR that implements the Harris corner detector, SIFT detector and descriptor, and two modifications of those algorithms specialized in HDR images, called SIFT for HDR (SfHDR) and Harris for HDR (HfHDR). Previous studies investigated the use of HDR images in FP detection, but we did not find studies investigating the use of HDR images in FP description. Using uniformity, repeatability rate, mean average precision, and matching rate metrics, we compared the performance of the CP_HDR algorithms using LDR and HDR images. We observed an increase in the uniformity of the distribution of FPs among the high-light, mid-light, and low-light areas of the images. The results show that using HDR images as input to detection algorithms improves performance and that SfHDR and HfHDR enhance FP description.

Figures (9)

• Figure 1: Selected and not selected studies in each of the search sources.
• Figure 2: Capture sequences examples from Přibyl et al. (2012) pvribyl2012 2D dataset. (a) and (b) from viewpoint capture sequence, (c) and (d) from distance capture sequence, and (e) and (f) from illumination capture sequence.
• Figure 3: Capture examples from Rana et al. (2015) rana2015 dataset. (a), (b) and (c) from ProjectRoom scene, and (d), (e), and (f) from LightRoom scene.
• Figure 4: Modification of the canonical detectors to implement the (a) Harris for HDR and (b) SIFT for HDR.
• Figure 5: Example of CV filter application on images: (a) reference image; (b) CV filter applied on a LDR capture; and (c) CV filter applied in a HDR capture.