High dynamic range imaging using vector space projection

The purpose of HDR imaging is to increase the dynamic range of the image intensity of a scene which otherwise would be limited by a normal imaging technique/device. A commonly used approach to HDR is to acquire images with multiple exposures of the same scene. In this paper, an iterative algorithm is explored for the first time to recover an HDR image from multiple exposures based on the method of projections onto constraint sets. To construct appropriate constraint sets, noise sources in pixels and sensor structures are discussed. An equivalent noise characteristics at the sensor output is modeled as the Gaussian Random Process. Base on noise characteristics and sensor response features, a group of pixel-wise constraint sets are defined to measure the estimation errors. Utilizing the knowledge on neighbourhood pixels, the local spatial constraint sets on the image are also defined. The method of vector-space projections is applied onto these constraint sets to reconstruct the HDR image using multiple exposures of the same scene.