Subpixel registration

Abstract

Sub pixel image alignment includes mapping first pixels from a first image and second pixels from a second image to a coordinate system and applying one or more sub-pixel shifts to the mapped first pixels. For each sub-pixel shift, an overall energy is calculated and is based on a plurality of gradients that represent changes in a channel value among the shifted first pixels and the mapped secondpixels. The sub-pixel alignment further includes determining the sub-pixel shift that provides the lowest overall energy.

Description

technical field [0001] The present invention relates to image registration technology, more particularly to a sub-pixel registration method and system. Background technique [0002] Robust image registration is one of the core problems of research in computer vision and the emerging field of computational photography (CP). Various robust image registration techniques now exist, and they are based on methods such as aligning features within an image or applying dense pixel-to-pixel correspondences. [0003] A general approach in computational photography is to take multiple photos of the same scene, each shot in a slightly different way, and then combine the images to produce a single photo. Some purposes for producing this final photo include, for example, refocusing the image or increasing the dynamic range of the photo in post-processing. In some cases, the purpose of CP is to produce a single super-resolution image based on multiple lower-resolution images. [0004] Al...

AI Technical Summary

Problems solved by technology

[0004] Although super-resolution images can be obtained by increasing the resolution of camera image sensors, higher resolution sensors also lead to higher manufacturing costs and greater shot noise

Also, increasing sensor chip size to accommodate a larger number of pixels results in increased capacitance, which reduces data transfer rates

Additionally, applying pixel-level registration between images during super-resolution image formation can lead to artifacts such as blurring and double features

Images