TY - CHAP
T1 - Robust registration of histological images
T2 - Techniques and applications
AU - Hsu, Wei Yen
PY - 2012/9
Y1 - 2012/9
N2 - Image registration that is used to reconstruct 3D structure of tissues from a series of images is an important topic in medical image analysis. However, it becomes a difficult challenge for image registration due to a variety of inherent factors, such as color difference and geometry discrepancy. In this chapter, a robust registration algorithm is proposed to automatically reconstruct 3D volume data from histological images. It mainly contains three procedures, including wavelet-based feature extraction, analytic robust point matching (ARPM), and registration refinement by modified Levenberg-Marquardt algorithm (FMLM). The concept that features could exist in multiscale is used to extract true feature points. The ARPM registration algorithm is then proposed to speedily accomplish the registration of two point sets with different size by simultaneously evaluating the spatial correspondence and geometrical transformation. Finally, a FMLM method is used to further refine registration results and achieve subpixel accuracy. The performance of proposed method is evaluated in comparison with several well-known approaches. The results indicate that the proposed method is a robust and fast method in image registration. registration results and achieve subpixel accuracy. The performance of proposed method is evaluated in comparison with several well-known approaches. The results indicate that the proposed method is a robust and fast method in image registration.
AB - Image registration that is used to reconstruct 3D structure of tissues from a series of images is an important topic in medical image analysis. However, it becomes a difficult challenge for image registration due to a variety of inherent factors, such as color difference and geometry discrepancy. In this chapter, a robust registration algorithm is proposed to automatically reconstruct 3D volume data from histological images. It mainly contains three procedures, including wavelet-based feature extraction, analytic robust point matching (ARPM), and registration refinement by modified Levenberg-Marquardt algorithm (FMLM). The concept that features could exist in multiscale is used to extract true feature points. The ARPM registration algorithm is then proposed to speedily accomplish the registration of two point sets with different size by simultaneously evaluating the spatial correspondence and geometrical transformation. Finally, a FMLM method is used to further refine registration results and achieve subpixel accuracy. The performance of proposed method is evaluated in comparison with several well-known approaches. The results indicate that the proposed method is a robust and fast method in image registration. registration results and achieve subpixel accuracy. The performance of proposed method is evaluated in comparison with several well-known approaches. The results indicate that the proposed method is a robust and fast method in image registration.
KW - Analytic robust point matching
KW - Medical image registration
KW - Modified Levenberg-Marquardt algorithm
KW - Neuro-informatics
KW - Spatial correspondence
UR - http://www.scopus.com/inward/record.url?scp=84892842610&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84892842610&partnerID=8YFLogxK
M3 - Chapter
AN - SCOPUS:84892842610
SN - 9781620810491
SP - 17
EP - 38
BT - Medical Imaging: Procedures, Techniques and Applications
PB - Nova Science Publishers Inc
ER -