Non-Rigid Registration of Fish Otolith Images

K. Nasreddine, A. Benzinou, and R. Fablet (France)


Otolith images, image matching, robustness, dynamic programming, iterative methods.


This paper deals with the automation of otolith image anal ysis. Previous works in the field of otolith imaging focused on ring detection, segmentation issues and the otolith nu cleus detection. Here, a method for matching otolith im ages is developped as a calibration tool that allows to com pare elementary features (structural/chemical) of an otolith set. As otolith images are poorly contrasted, intensity based registration fails in aligning ring structures, while the proposed geometric registration is able to align both geometry and gray intensity of images. A variational prob lem that minimizes the cost of shape deformation is con sidered; the taken shapes are those of constructed potential functions that model the variation of the otolith geometry throughout the fish life. This problem of geometric reg istration is solved using many methods; the comparison of their results shows the superiority of a robust iterative scheme. Significant improvements are reported about this geometric registration of otolith images compared to iconic one. Besides its providing us to do symmetric registration, the proposed metric can be used to do statistical studies on the otolith shapes.

Important Links:

Go Back