Detecção e descrição de pontos de controle em imagens HDR

Abstract

In computer vision, a characteristic refers to image regions with unique properties, such as corners, edges, textures, or areas with high contrast. These regions are called control points (CPs). CP detectors and descriptors identify features in images and are the basis for many applications, such as object recognition, three-dimensional scene reconstruction, and biometric systems. Most CP detection and description methods use low dynamic range (LDR) images, which are sufficient for most applications involving with digital images. However, this type of representation limits the dynamic range and does not properly represent light under extreme lighting conditions. High dynamic range (HDR) images allow the representation of a wider range of lighting intensities. Consequently, lighting extremes are better represented in HDR images. This work investigates the potential of using HDR images in CP detectors and descriptors. We developed the CP_HDR library that implements Harris and Harris for HDR detection algorithms and SIFT and SIFT for HDR detection and description algorithms. Using uniformity, repeatability rate, mean average precision, and matching rate metrics, we compared the performance of the CP_HDR algorithms. We observed that when using HDR images with specialized detectors for HDR images, there is an increase in the distribution of CPs detected in the darkest, medium, and brightest areas of the images. We also observed that the description produced using HDR images with the canonical algorithms provided a better description of the CPs. The results show that using HDR images as input in detection algorithms improves its performance, and the implemented algorithms specialized for HDR images enhance the description of CPs.