Raw camera noise reduction using alignment mapping

Abstract

This disclosure pertains to novel devices, methods, and computer readable media for performing raw camera noise reduction using a novel, so-called 'alignment mapping' technique to more effectively separate structure from noise in an image, in order to aid in the denoising process. Alignment mapping allows for the extraction of more structure from the image and also the ability to understand the image structure, yielding information for edge direction, edge length, and corner locations within the image. This information can be used to smooth long edges properly and to prevent tight image details, e.g., text, from being overly smoothed. In alignment maps, the amount of noise may be used to compute thresholds and scaling parameters used in the preparation of the alignment map. According to some embodiments, a feature map may also be created for the image. Finally, the image may be smoothed using the created feature map as a mask.