Premium
Analytic differential approach for robust registration of rat brain histological images
Author(s) -
Hsu WeiYen
Publication year - 2011
Publication title -
microscopy research and technique
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.536
H-Index - 118
eISSN - 1097-0029
pISSN - 1059-910X
DOI - 10.1002/jemt.20942
Subject(s) - image registration , artificial intelligence , computer science , iterative closest point , subpixel rendering , pattern recognition (psychology) , point set registration , feature (linguistics) , computer vision , transformation (genetics) , matching (statistics) , feature extraction , spline (mechanical) , robustness (evolution) , point (geometry) , image (mathematics) , mathematics , point cloud , pixel , linguistics , philosophy , biochemistry , geometry , chemistry , statistics , structural engineering , gene , engineering
Image registration is an important topic in medical image analysis. It is usually used to reconstruct 3D structure of tissues from a series of microscopic images. However, a variety of inherent factors may result in great differences between acquired slices during imaging even if they are adjacent. The common differences include the color difference and geometry discrepancy, which make the registration problem a difficult challenge. In this study, we propose a robust registration method to automatically reconstruct 3D volume data of the rat brain. It mainly consists of three procedures, including multiscale wavelet‐based feature extraction, analytic robust point matching (ARPM), and registration refinement with feature‐based modified Levenberg‐Marquardt algorithm (FMLM). The product of gradient moduli in multi‐scales is used to decide if extracted feature points are true according to the characteristic that features could exist in multiscale. The ARPM registration algorithm is proposed to speedily accomplish the registration of two point sets with different size by simultaneously evaluating the spatial correspondence and geometrical transformation. In addition, a FMLM method is also proposed to further refine registration results and achieve subpixel accuracy. The FMLM method converges much faster than most other methods due to its feature‐based and nonlinear characteristic. The performance of proposed method is evaluated by comparing it with well‐known thin‐plate spline robust point matching (TPS‐RPM) algorithm. The results indicate that ARPM‐FMLM algorithm is a robust and fast method in image registration. Microsc. Res. Tech., 2010. © 2010 Wiley‐Liss, Inc.