Interferometric lens-free microscopy of structured materials

The lens-free interferometric microscope (?LIM?) developed at ICFO by Roland Terborg et al. offers a hybrid solution of differential interference contrast microscopy, interferometry and holography. It is an ultra-sensitive interferometric microscope used for the analysis of structures and defects on the surface or inside the bulk of transparent materials, that can be both in transmission and reflection. It is an evolving platform, both from a hardware and software perspective. One of the main tasks in this thesis was to add an image processing algorithm to the LIM platform that can be used to detect, mark and analyse optical phase features on or inside transparent samples. The segmentation algorithm, based on edge detection, watershed transform and K-means classification, is used to classify phase maps into background and feature. Once classified into a binary map, connected areas are assigned to an integer label and the properties of the underlying phase map are measured. This allows to perform a systematic quantitative and statistical analysis of phase features by discarding the information that is out of focus. When samples containing features of known size are measured, the algorithm can be tailored for a detailed analysis of these feature structures, not only in 2D but also in 3D. The segmentation and labelling algorithm is augmented into a labelling based autofocus, that is capable of finding features planes or layers with known geometric features. Several tests are performed to prove the capabilities of the segmentation and labelling algorithm both in 2D and 3D, in combination with the proposed autofocus. Part of this is an analysis of embedded refractive index structures inside a 1 mm thick glass slide. The structures are completely invisible to the human eye and other microscopes. The smallest features are 50 by 50 µm in size and induce an optical distortion equivalent to 10 nm on the wavefront of transmitted light. The LIM is capable of making them visible. The features can be segmented and analysed and the algorithm can find and reconstruct layers separated by only 450 µm