A modified collaborative adaptive wiener filter for multi-frame super-resolutionaper

During acquisition, digital images are invariably degraded by a number of phenomena that limit image resolution and utility. Aliasing from undersampling, blur from optics, and sensor noise are some factors which can affect the image resolution. Multi-frame super-resolution (SR) is a technique that takes several low-resolution (LR) frames of a particular scene and processes them together to produce one or more high-resolution (HR) images. The HR images have higher spatial frequency content, and less noise and blur, than any of the LR frames. A collaborative adaptive Wiener filter (CAWF) for multi-frame SR, proposed by the current authors, is one of the very recent effective multi-frame SR algorithms. In this paper, we modify the original CAWF SR method by employing a spatially varying signal variance estimate. Instead of using a global signal variance estimate as an external input to the original CAWF SR algorithm, we estimate the desired signal variance in each processing window and incorporate it to estimate the HR pixels. The modified CAWF SR is presented and demonstrated. In addition, performance comparisons between the original and the modified CAWF SR are conducted. The modified CAWF SR outperforms the original CAWF SR, particularly in low signal-to-noise ratio images.