IMAGE REGISTRATION

• Classification
  Nature: area-based and feature-based.

Different viewpoints (multiview analysis): image from different viewpoints, 2D or 3D representation like remote sensing and stereo vision.

Different times (multitemporal analysis): images at different times, in consecutive acquisitions. Automatic change detection for security monitoring, motion tracking.

Different sensors (multimodal analysis): images from different sensors, obtained from different source to represent more details. Remote sensing with better resolution. Radar images independent of cloud cover.

Scene to model registration: images of a scene and a model of the scene are registered. To compare similar images. Target template matching with real-time images.

• Steps

  1. feature detection: edge, contour, line intersection, corners, closed-boundary. Called control points(CP).
  2. feature matching: various feature descriptors and similarity measures along spatial relationships.
  3. transform model estimation: mapping function computed by means of established feature correspondence.
  4. image resampling and transformation: sensed image transformed by means of mapping functions. Else is computed by interpolation techniques.

• feature detection
  feature choice is important according to task.
  robust to incorrect feature detection or image degradations.
  

  • area-based methods: omit this step.
  • feature-based methods: significant regions、lines（region boundaries、coastlines）、points（region corners、line intersections）
    

  region features(high contrast closed-boundary. Representing by gravity. Detected by segmentation methods.)
  
  N.R. Pal, S.K. Pal, A review on image segmentation techniques, Pattern Recognition 26 (1993) 1277–1294
  
  A. Goshtasby, G.C. Stockman, C.V. Page, A region-based approach to digital image registration with subpixel accuracy, IEEE Transactions on Geoscience and Remote Sensing 24 (1986)390–399.
  
  A. Noble, Finding corners, Image and Vision Computing 6 (1988) 121–128.
  
  As for line feature, line segmentation, object contours、coastline、roads、structure.
  
  Y.C. Hsieh, D.M. McKeown, F.P. Perlant, Performance evaluation of scene registration and stereo matching for cartographic feature extraction, IEEE Transactions on Pattern Analysis and Machine Intelligence 14 (1992) 214–237
  
  J. Canny, A computational approach to edge detection, IEEE Transactions on Pattern Analysis and Machine Intelligence 8 (1986) 679–698.
  
  Point features: line intersections、roads crossing、centroids、oil、gas pads.
  
  B. Likar, F. Pernus, Automatic extraction of corresponding points for the registration of medical images, Medical Physics 26 (1999) 1678–1686.
  
  D. Bhattacharya, S. Sinha, Invariance of stereo images via theory of complex moments, Pattern Recognition 30 (1997) 1373–1386.
  
  T.M. Lehmann, A two stage algorithm for model-based registration of medical images, Proceedings of the Interantional Conference on Pattern Recognition ICPR’98, Brisbane, Australia, 1998, pp. 344–352.(corner)
  
  P. Hellier, C. Barillot, Coupling dense and landmark-based approaches for non rigid registration, IRISA research report, PI 1368:30, France, 2000.
  
  All about feature-based methods are effective when the objects are detectable.

• feature matching

  1. area-based methods: correlation-like methods or template matching. Combining the detection step with matching part. Deformation maybe complex to recover.

     L.M.G. Fonseca, B.S. Manjunath, Registration techniques for multisensor remotely sensed imagery, Photogrammetric Engineering and Remote Sensing 62 (1996) 1049–1056.

     W.K. Pratt, Correlation techniques of image registration, IEEETransactions on Aerospace and Electronic Systems 10 (1974)353–358.

  2. feature-based methods:

• transform model estimation
  reserve useful information and keep change.

• image resampling and transformation
  trade-off between accuracy and computational complexity. The nearest-neighbor or bilinear interpolation are sufficient.

• To be continued……