|Title||Alignment of confocal scanning laser ophthalmoscopy photoreceptor images at different polarizations using complex phase relationships|
|Publication Type||Journal Article|
|Year of Publication||2010|
|Journal||IEEE Transactions on Biomedical Engineering|
|Keywords||Algorithms, Animals, bio-optics, biomedical optical imaging, complex phase relationship, Computer-Assisted, Confocal, confocal scanning laser ophthalmoscopy, enhancing fundus image, eye, Fishes, Fundus Oculi, image enhancement, image processing, incoming polarization states, laser applications in medicine, light polarisation, medical image processing, Microscopy, Ophthalmoscopy, Photoreceptor Cells, photoreceptor image, polarimetry, Polarization, quadratic programming, sequential quadratic programming, spatially resolved Mueller image construction, Thermodynamics, Vertebrate|
A polarimetric technique for enhancing fundus images was recently introduced , where confocal scanning laser ophthalmoscopy (CSLO) images are acquired under differing incoming polarization states, and spatially resolved Mueller images are constructed based on the images. An important stage in this technique is the alignment of CSLO images acquired under differing polarization states. This has proven to be particularly difficult when dealing with photoreceptor images, which are characterized by poor SNRs and intensity inhomogeneities due to polarization properties. In this paper, an automated approach to aligning CSLO photoreceptor images acquired under differing polarization states is presented. A novel energy functional based on complex phase relationships is introduced that is invariant to polarization and scale, as well as robust to noise and highly sensitive to photoreceptor structural characteristics. A sequential quadratic programming approach is employed to determine the optimal alignment between the photoreceptor images by minimizing the proposed energy functional. The method has been tested on CSLO fish photoreceptor images acquired under differing polarization states and evaluated based on alignment accuracy. The results demonstrate that the proposed method outperforms existing techniques used for aligning CSLO images, with lower mean alignment error for all test cases.