PROJECT TITLE :

All-Pass Parametric Image Registration

ABSTRACT:

Many practical applications involving the capture of numerous linked images necessitate image registration. In this paper, we present a strategy for dealing with the image registration problem's geometric and intensity transformations. An elastic registration procedure (Local All-Pass-LAP) is modified to return a parametric displacement field, and the intensity changes are estimated by fitting another parametric expression to the returned displacement field. A low-order parametric model is used to explain the process, but it is highly adaptable and may be used to create far more complex parametric models with minimal additional computing cost. The "salience correlation" and the "parsimony" metrics, which measure the number of degrees of freedom in a displacement field, are new quantitative criteria we've developed to assess alignment precision. The accuracy and computational efficiency of our method have been demonstrated in both synthetic and real-world photographs through the use of experimental data. As an added benefit, our results show that the resulting displacement fields are more sparse than those obtained by other current picture registration methods.