Pixel-wise registration of close-range multispectral images acquired with multi-lens cameras

Abstract. Lightweight multispectral cameras have been widely used for many applications, mainly agriculture. Suitable results in image analysis processes depend on the accurate registration of the image bands, collected by lenses that are mutually displaced. The problem of registering image bands of close-range objects collected by multispectral cameras is still lacking in effective solutions. Most of the existing approaches are not capable of accurately matching each pixel in the images since depth variations still cause displacements. In this paper, we proposed a solution by extracting a digital surface model (DSM) and generating orthoimages of each image band. We experimentally showed that image bands of the same frame shot, collected at the same station and time, can be used to generate a DSM and orthoimages, which will be mutually registered. To achieve this, a rigorous camera calibration must be previously performed to provide the orientation parameters for the DSM extraction and orthorectification. The experimental results showed that the standard deviation of the orthorectified pixels is approximately 1.5 pixels.