In this paper, an irregular displacement-based lensless wide-field microscopy imaging platform is presented by combining digital in-line holography and computational pixel super-resolution using multi-frame processing. The samples are illuminated by a nearly coherent illumination system, where the hologram shadows are projected into a complementary metal-oxide semiconductor-based imaging sensor. To increase the resolution, a multi-frame pixel resolution approach is employed to produce a single holographic image from multiple frame observations of the scene, with small planar displacements. Displacements are resolved by a hybrid approach: (i) alignment of the LR images by a fast feature-based registration method, and (ii) fine adjustment of the sub-pixel information using a continuous optimization approach designed to find the global optimum solution. Numerical method for phase-retrieval is applied to decode the signal and reconstruct the morphological details of the analyzed sample. The presented approach was evaluated with various biological samples including sperm and platelets, whose dimensions are in the order of a few microns. The obtained results demonstrate a spatial resolution of 1.55 μm on a field-of-view of ≈30 mm2.
4 January 2021 in Pattern Analysis and Applications
Real-time human facial expression recognition plays a significant role in many application areas, including human-computer interaction, business intelligence, video surveillance, and robotics. Based on facial expressions, computers can interpret human feelings and psychological stages to pro- vide more realistic approximations in real-world applications. This paper proposes a simple but effective solution for real-time Facial Emotion Recognition (FER), using a mask of the most relevant facial features as input data for a machine-learning approach. For this, a compact Con- volutional Neural Network (CNN) classifier associated with a feature extraction layer was used to provide an end-to-end solution that can detect facial expressions from videos with good accuracy rates. The pro- posed approach was validated using a combination of different facial emotion datasets available in the literature, whose precision rates are considerably better than those provided by the state-of-the-art methods. Score rates of 96.83%, 98.58%, and 98.57% were obtained for the JAFFE, RaFD, and CK+ datasets, respectively, indicating that the presented approach is a promising solution for FER in real-time applications. DOI:https://doi.org/10.56238/sevened2024.004-006