A Novel Local Invariant Feature Extraction Method for High-dynamic Range Images

Abstract

In order to perform target recognition under realistic conditions, it is necessary to extract local invariant features of a real scene. This paper proposes a novel local invariant feature extraction method for high dynamic range images. Firstly, based on the multi-exposure fusion HDR image calculation model, this paper introduces the perceptual model of human eye vision for scene irradiance, and establishes a complete theoretical model of HDR imaging and mapping. Then, this paper proposes a new LIFT extraction method, which performs LIFT detection on the scene reflection layer and LIFT description on the scene illumination layer. Finally, experiments have shown that this method can increase the number of correct matching of feature points of machine vision wide baseline images. The research results in this paper combine the invariant feature matching technology with the HDR image technology that can record the irradiance value of the real scene, thus expanding the application background of the image LIFT extraction technology.