Image processing methods utilizing motion information: Dissertation

This thesis concerns the development of methods and algorithms for image sequence analysis.The study covers a wide range.However, motion as a common feature runs through the whole thesis.The following topics have been covered in the thesis: motion estimation from an image sequence, determination of depth from a zooming image sequence, estimation of 3-D parameters, moving object recognition and image sequence coding.Motion estimation from an image sequence is considered and a 2-D displacement vector field is estimated for various applications.A new motion constraint equation is presented to avoid the problem caused by non-uniform illumination.A new additional constraint is introduced to preserve the discontinuities in the estimation.The estimated 2-D displacement vector field is successfully applied to image segmentation.A zooming camera is considered for determination of surface depth in a static scene.Camera zooming can be regarded as a special type of motion and depth can be recovered from the zooming image sequence.Two schemes are discussed.Image irradiance equations for a zooming camera are developed.The estimation of 3-D parameters from a 2-D image sequence is studied.An algorithm which uses only three straight line correspondences over two frames is presented.Image recognition for moving objects is considered.A fast shape descriptor is presented and a relaxation labeling based edge detector is developed.This fast shape descriptor has some advantages over the Fourier descriptors.The fast shape descriptor is applied to airplane tracking.Two schemes for image compression are investigated.The first-scheme presents a novel idea for image sequence coding and it emphasizes the use of temporal derivatives.The second scheme discusses a fractal based image compression method.The images are split into a group of small squares according to its fractal dimension instead of contour.