Frequency selective height-from-stereo reconstruction: Application to retinal oedemas detection

The wavelet transform decomposes a given signal into a time-frequency representation, preserving the locality of both. Given that the local relief elevation is correlated to the phase difference in the stereo pair of images, we demonstrate a frequency selective relief reconstruction from a stereo pair of images, utilizing the dual-tree complex wavelet transform to provide phase information. The dual-tree complex wavelet decomposition enables the hierarchical measurement of the phase difference in situations where the relief elevation is only observable in limited components of the image. Incorporating a data specific image registration technique, this paper presents an new application of the reconstruction of the macular oedema from a stereo pair of eye fundus photographs, offering the potential to serve as a diagnostic aid in the process of detecting macular oedema.