Iterative point matching for registration of free-form curves and surfaces
International Journal of Computer Vision
Pattern Classification (2nd Edition)
Pattern Classification (2nd Edition)
Image Processing, Analysis, and Machine Vision
Image Processing, Analysis, and Machine Vision
Hi-index | 0.00 |
One way of evaluating muscle quality is to detennine its fiber type composition in histological sections. A complete muscle fiber type characterization system requires combining information from successive muscle histology images with different ATPase stain. Due to the local and global defonnations introduced in slide preparation process, a precise non-rigid registration is essential to construct the spatial correspondences between these successive images. This study proposes an approach for automated non-rigid registration of successive muscle histological sections. We propose a feature-based registration that uses a two stage approach: a rigid initialization followed by a non-rigid refinement. The rigid initialization step globally aligns successive tissue slides by finding correspondences between individually segmented muscle fibers using Fourier shape descriptors and computing the global rigid transformation using a voting scheme tolerant to mismatches. In the non-rigid stage we establish precise point correspondences using the nonnalized cross correlation metric and compute the non-rigid distortion using a polynomial transformation that minimizes the mean square distance between these control points.