Abstract
Digital Lensless Holographic Microscopy (DLHM) presents a powerful methodology for in-situ monitoring and characterization of underwater microorganisms, offering valuable insights into marine ecosystems. However, the widespread adoption of DLHM in oceanographic research has been constrained by the computational complexity and limitations of existing hologram reconstruction frameworks. This Master’s Thesis investigates the development of an accurate, robust, and computationally efficient framework designed to overcome these challenges using data acquired from the Deep-See DLHM system. The proposed framework addresses critical deficiencies in current approaches by incorporating an additional step for speckle noise reduction, modifying the method for uneven illumination correction using polynomial fitting, and implementing a different segmentation algorithm that surpasses the limitations of simpler circle-detection algorithms, which present a poor performance for large microorganisms, improving detection rates while substantially reducing computational time compared to previous methods. Unlike traditional imaging techniques, the DLHM system records defocused images of the targeted organisms, which requires that the recorded image for each segmented microorganism must be computationally backpropagated using the angular spectrum method to provide an in-focus image. In this Master’s Thesis, we also investigate the behavior of multiple reported sharpness metrics to automatically provide in-focus amplitude images for five different types of marine planktonic organisms. Our results demonstrate that the Tenegrand variance metric is the most robust, consistently presenting a maximum peak when these various plankton organisms are in focus. The search for the in-focus plane for each segmented microorganism has been implemented using the traditional conventional grid search in which the Tenegrand variance metric is evaluated for each reconstructed amplitude distribution. Finally, to reduce the computational time of the entire process by a factor of 1.5, the focusing algorithm has been implemented using a two-step approach: a first coarse search to determine the potential distance where a microorganism is located and a second fine-step search to accurately determine its location. In conclusion, the developed computational DLHM framework improves both the accuracy and computational efficiency in detecting and focusing individual microorganisms within a large field of view, providing a more effective and comprehensive tool for marine ecological research.