Analysis of medical image sequences by recursive polynomial registration

This paper presents a new technique of polynomial image registration, applying recursive least squares techniques for the estimation of the mapping function coefficients. The described technique can be used with efficiency for registration of images of non-rigid bodies with low frequency local variations. In addition, the incorporation of weights for the samples is foreseen, reflecting the visual trust in the acquired control points. The technique is applied in the construction of a system for aid in the differential diagnosis of lung diseases, with low computational cost and larger interaction with the operator, since the recursive nature of the algorithm facilitates the update of the registration during the acquisition of the control points.