Multiplexed image reconstruction, estimating high resolution images from multiple low resolution images with highly overlapped fields of view, is improved when the magnification of the imagers is diverse. No assumptions of shift invariance or Toeplitz structure are required for computational manageability because localized reconstruction is possible and sensitivity to boundary conditions is reduced. Such multiplexed diverse image sensors have applications in flat sensor systems for surveillance and pervasive personal imaging.

Previous work has shown that for super-resolution image reconstruction from low resolution images, image acquisition with a diverse optical system improves reconstructed image quality as measured by the expected and actual mean squared error. However, other measures of image fidelity should also be considered. An alternative performance measure might be based on edge errors, since edges are often the first step in more complex image analysis for both image processing systems and biological systems. This paper explores the behavior of edge errors and intensity errors for super-resolution image reconstruction applications in which ill-posed inversions may cause the actual mean squared error to be highly dependent of image content and thus poorly predicted by the expected mean squared error.

Abstract-High resolution images can be estimated using multiple low resolution images obtained from an array of subimagers with overlapping fields of view. Design choices for the optics and sensors of a flat camera can have a significant impact on the performance of reconstruction algorithms. This paper will analyze designs that provide diversity which reduces the expected error of reconstruction algorithms. It will be assumed that the fields of view of individual sub-imagers can be controlled so that the desired resolution improvement at the object distance of interest can be achieved. Simulations will demonstrate the analytical results.

No abstract