Mozhdeh Seifi

Fast and accurate 3D object recognition directly from digital holograms

Pattern recognition methods can be used in the context of digital holography to perform the task of object detection, classification, and position extraction directly from the hologram rather than from the reconstructed optical field. These approaches may exploit the differences between the holographic signatures of objects coming from distinct object classes and/or different depth positions. Direct matching of diffraction patterns, however, becomes computationally intractable with increasing variability of objects due to the very high dimensionality of the dictionary of all reference diffraction patterns. We show that most of the diffraction pattern variability can be captured in a lower dimensional space. Good performance for object recognition and localization is demonstrated at a reduced computational cost using a low-dimensional dictionary. The principle of the method is illustrated on a digit recognition problem and on a video of experimental holograms of particles.

Three-dimensional reconstruction of particle holograms: a fast and accurate multiscale approach.

In-line digital holography is an imaging technique that is being increasingly used for studying three-dimensional flows. It has been previously shown that very accurate reconstructions of objects could be achieved with the use of an inverse problem framework. Such approaches, however, suffer from higher computational times compared to less accurate conventional reconstructions based on hologram backpropagation. To overcome this computational issue, we propose a coarse-to-fine multiscale approach to strongly reduce the algorithm complexity. We illustrate that an accuracy comparable to that of state-of-the-art methods can be reached while accelerating parameter-space scanning.

Inverse problems approaches for digital hologram reconstruction

Digital holography (DH) is being increasingly used for its time-resolved three-dimensional (3-D) imaging capabilities. A 3-D volume can be numerically reconstructed from a single 2-D hologram. Applications of DH range from experimental mechanics, biology, and fluid dynamics. Improvement and characterization of the 3-D reconstruction algorithms is a current issue. Over the past decade, numerous algorithms for the analysis of holograms have been proposed. They are mostly based on a common approach to hologram processing: digital reconstruction based on the simulation of hologram diffraction. They suffer from artifacts intrinsic to holography: twin-image contamination of the reconstructed images, image distortions for objects located close to the hologram borders. The analysis of the reconstructed planes is therefore limited by these defects. In contrast to this approach, the inverse problems perspective does not transform the hologram but performs object detection and location by matching a model of the hologram. Information is thus extracted from the hologram in an optimal way, leading to two essential results: an improvement of the axial accuracy and the capability to extend the reconstructed field beyond the physical limit of the sensor size (out-of-field reconstruction). These improvements come at the cost of an increase of the computational load compared to (typically non iterative) classical approaches.

Dictionary size reduction for a faster object recognition in digital holography

Pattern matching methods can be used in the context of digital holography to perform the task of object recognition, classification and position extraction directly from the hologram and not from the reconstructed optical yield. These approaches exploit the differences between the objects holographic signatures caused by class and depth position of the objects. In this talk we will show that such inter-signature variabilities can be captured efficiently in a lower-dimensional vector space using dimensionality reduction methods.