M.Sc Thesis

M.Sc StudentLitvinov Anatoly
SubjectImage Mosaicing in the Presence of Radiometric Distortions
DepartmentDepartment of Electrical and Computer Engineering
Supervisor PROF. Yoav Schechner


Cameras may have non-ideal radiometric aspects, including spatial non-uniformity, e.g., due to vignetting or foreshortening; a non-linear radiometric response of the camera; and temporal variations due to automatic gain control. Often, these characteristics exist simultaneously, and are typically unknown. These characteristics hinder photometric measurements. They are particularly annoying in image mosaicing, where images are stitched to enhance the field of view. Mosaics suffer from seams stemming from radiometric inconsistencies between raw images. Prior studies approached part of these radiometric problems while excluding others. In addition, several methods were developed to feather seams in image mosaics or camouflage them. However, these mosaicing methods do not address their root causes of the mismatch. In this work all the above mentioned radiometric problems are handled in a unified framework. A method is suggested for simultaneously estimating the radiometric response, the camera non-uniformity and frame gain, based on a frame sequence acquired during camera motion. Our approach does not rely on dedicated processes that intentionally vary exposure settings and does not assume parametric models to the radiometric function or spatial non-uniformity. We define the fundamental equations and formulate the problem as a least mean square problem. Then we suggest a simple method to solve it in matrix form. Fundamental ambiguities and trivial solutions associated with this estimation problem are stated. We then describe simple methods to overcome them. In addition, a novel image invariance is exposed. The estimated functions are then compensated for. This enables image mosaicing, in which no seams are apparent. We demonstrate our method in several experiments, where different frames are brought into mutual radiometric consistency. In these experiments, the compensation was good enough to alleviate the need for to dedicated seam-feathering methods. Finally, we propose options to extent our research.