Robust image alignment for distributed multi-view imaging systems

摘要

Disclosed is a method of image coding for joint decoding of images from different viewpoints using distributed coding techniques. The method receives a first set of features (205) and error correction bits (203) corresponding to a first image (201) obtained at a first viewpoint (122) and a second set of features (425) from a second image (254, 415) corresponding to a second viewpoint (124). An approximation (437) of said first image (201) at said first viewpoint (122) is determined (432, 434, 436) an based on the first and second sets of features (205, 425) and the second image at the second viewpoint. A reliability measure (445) of the approximation of the first image is then determined (450) by joint decoding (438) the approximation (437) using the error correction bits (203). The approximation of the first image is then refined iteratively (460, 438) based on the reliability measure (445) and image information (448) derived from the joint decoding.

主权项

1. A method of joint decoding of images from different viewpoints using distributed coding techniques, said method comprising the steps of:
receiving a first set of features from a first image and error correction bits corresponding to said first image, said first image being obtained at a first viewpoint; receiving a second image obtained at a second viewpoint and a second set of features from said second image corresponding to said second viewpoint, said second viewpoint being different from said first viewpoint; determining an approximation of said first image at said first viewpoint based on said first and second sets of features and said second image at said second viewpoint; jointly decoding said approximation of said first image using said error correction bits; determining a reliability measure of said decoded approximation of said first image, the reliability measure being an indication of a confidence of values of bits of the decoded approximation; refining said approximation of said first image iteratively based on said first and second sets of features and said second image and said decoded approximation if the reliability measure is below a predetermined threshold, wherein said approximation of said first image is iteratively refined until the reliability measure exceeds the predetermined threshold; and outputting said decoded approximation as a decoded image where the reliability measure exceeds the predetermined threshold.