Dr. Nelson Díaz

Dr. Nelson Díaz

Nelson Díaz, contact information: nelson [dot] diazd [at] usm [dot] cl

Research Statement

Published:

My computational imaging: laboratory will focus on designing and developing intelligent sensing architectures and recovery algorithms that improves our understanding of the world by enabling more efficient acquisition, representation, and interpretation of high-dimensional light field (space, time, depth, wavelength, polarization, etc).


Research Thrusts

My research in computational imaging: spans multiple interconnected areas of computer science, such as, data science, machine learning, computer vision, algorithm design, linear algebra, numerical analysis, mathematical modeling, software, and inverse problems. In addition, my research areas require skills in electronic engineering, such as digital signal/image processing, instrumentation, optical multidimensional data sampling, and compressive sensing. I learned these competencies during my bachelor's, master's, and doctoral studies, and I strengthened them during my postdoctoral position (ANID Fondecyt Postdoctorado 3230489).


My research thrusts can be grouped into three main research areas and are not restricted to: (1) High dimensional signal processing and compressive sensing (2) Algorithms design, inverse problems and machine learning. (3) Remote Sensing and Geoinformatics. These research thrusts will have a broad impact on several domains including, consumer photography, surveillance, remote sensing, smart agriculture or agriphotonics, target identification, medical applications, microscopy, computational imaging, and applied optics.


High Dimensional Signal Processing and compressive sensing:

Multispectral imaging and Demosaicking: I have designed optical coding for multispectral filter array with an optimal sensing distribution exploiting sphere packing that overcomes the limitations of traditional multispectral filters (color distortion, zipper effect, and color artifacts) [J5], that was published in The IEEE Transactions on image processing a journal with impact factor 15.3. Our approach is a general framework for spectral imaging acquisition, achieving notable image quality, independent of the reconstruction approach (classical interpolation, convex optimization, and novel neural networks).


Video sensing: I have developed a novel temporal coding using sphere packing approach for shuffled the scanlines of the rolling shutter of the CMOS camera, which eliminate the nuisance artifacts of the rolling shutter mechanism found in complementary metal oxide semiconductor (CMOS) detectors, Journal [J6,B2]. The temporal coding using single-mask sphere packing has been combined with implicit neural representation for ultrahigh-speed imaging, see Journal [J4,B1,C1].


Compressive spectral imaging: Furthermore, journals [J1, J8-J10, C6] show my expertise in compressive spectral imaging and adaptive sensing. Journal [J9] extends my master's work in high dynamic range compressive spectral imaging (CSI) using real data and using an optimal coded aperture according to restricted isometry property (RIP), such as blue noise patterns. Wavefront coding using diffractive optical elements for solving tasks such as multispectral extended-depth-of-field (EDoF) [J1].


Compressive video imaging: My skills with deep learning, ultrasound imaging, and temporal spectral video are shown in references [C2], [C3], and [C4], respectively. In [J2], we propose a novel and scalable sampling scheme for compressive video imaging. Our method is compatible with a wide range of temporal sampling matrices and reconstruction algorithms. It is capable of achieving image resolutions up to $2K \times 2K$ in a single snapshot, even at high compression ratios. I have also developed a novel coded aperture design for phase retrieval, as shown in [J1].


Algorithms design, inverse problem and machine learning: One key aspect in my research areas involves reconstruction methods by solving inverse problems using either iterative regularized algorithms [J1,J3, J8-J10, C5] or novel neural networks that are inspired in physics-informed neural representation approaches such as implicit neural representation [J4], convolutional neural network [J2][J5], [J6].


Remote Sensing and Geoinformatics: Two modalities are highlighted here, compressive spectral imaging classification, and point cloud processing captured using Light Detection and Ranging (LiDAR). On the one hand, I design adaptive compressive multispectral imaging and hyperspectral acquisition systems to improve the image quality and perform specific tasks, such as classification. Specifically, my work, published in [J8], combines a two-arm optical system, i.e, the CSI architecture includes one arm that samples the projection of multispectral data and the other arm that captures hyperspectral projections. An adaptive system is used to design the complementary coded aperture patterns to improve the spectral classification accuracy. On the other hand, I developed and implemented an algorithm for ground filtering using point cloud processing captured using Light Detection and Ranging (LiDAR). In detail, the journal in [J7] demonstrates my expertise in handling various types of multidimensional data, including point clouds acquired using LiDAR for specific tasks like ground filtering.


  1. [J1] Exequiel Oliva, Nelson Diaz, Samuel Pinilla, and Esteban Vera. Multispectral extended depth-of-field imaging via stochastic wavefront optimization. IEEE Open Journal of Signal Processing, 6:965–974, 2025, doi:https://doi.org/10.1109/OJSP.2025.3595046
  2. [J2] Felipe Guzmán, Nelson Diaz, Bastián Romero, and Esteban Vera. Scalable coding for high-resolution, high-compression ratio snapshot compressive video. IEEE Transactions on Image Processing, 34:3960– 3970, 2025, doi:https://doi.org/10.1109/TIP.2025.3579208
  3. [J3] Bastián Romero, Pablo Scherz, Nelson Diaz, Jorge Tapia, Aarón Cofré, Eduardo Peters, Esteban Vera, and Darío G. Pérez. Phase retrieval by designed hadamard complementary coded apertures. Optics Laser Technology, 191:113311, 2025, doi:10.1016/j.optlastec.2025.113311.
  4. [J4] Nelson Diaz, Madhu Beniwal, Miguel Marquez, Felipe Guzman, Cheng Jiang, Jinyang Liang, and Esteban Vera. Single-mask sphere-packing with implicit neural representation reconstruction for ultrahigh-speed imaging. Opt. Express, 33(11):24027–24038, Jun 2025, doi:10.1364/OE.561323.
  5. [J5] N. Diaz, A. Alvarado, P. Meza, F. Guzmán and E. Vera, Multispectral Filter Array Design by Optimal Sphere Packing,” in IEEE Transactions on Image Processing, vol. 32, pp. 3634-3649, 2023, doi:10.1109/TIP.2023.3288414.
  6. [J6] E. Vera, F. Guzman, N. Díaz, Shuffled Rolling Shutter for Snapshot Temporal Imaging, Opt. Express, 2022, doi.org/10.1364/OE.444864.
  7. [J7] N. Díaz, et al, Real-time ground filtering algorithm of cloud points acquired using Terrestrial Laser Scanner”, International JAG, 2021, doi.org/10.1016/j.jag.2021.102629.
  8. [J8] N. Díaz, Juan Ramirez, Esteban Vera, Henry Arguello. Adaptive multisensor acquisition via spatial contextual information for compressive spectral image classification, IEEE JSTARS, 2021, doi:10.1109/JSTARS.2021.3111508
  9. [J9] N. Díaz, Carlos Hinojosa, Henry Arguello, Adaptive grayscale compressive spectral imaging using optimal blue noise coding patterns, Optics \& Laser Technology, 2019, doi:10.1016/j.optlastec.2019.03.038.

  10. [J10] N. Díaz, H. Rueda, H. Arguello, Adaptive filter design via a gradient thresholding algorithm for compressive spectral imaging, App. Optics, 2018, doi:10.1364/AO.57.004890.

  11. [B1] Esteban Vera, Felipe Guzman, and Nelson Diaz. Shuffled Rolling Shutter Camera, pages 499–513. Springer International Publishing, Cham, 2024 doi:10.1007/978-3-031-39062-3_27

  12. [B2] James Skowronek, Felipe Guzmán, Nelson Diaz, Esteban Vera, and David Brady. Space-time imaging. Handbook of Statistics. Elsevier, 2026 doi:10.1016/bs.host.2026.03.004

  13. [C1] N. Diaz, M. Beniwal, F. Guzmán, M. Marquez, J. Liang, and E. Vera. “Binary Coded Aperture Design by Sphere Packing in Compressive Ultrafast Photography,” in Optica Sensing Congress 2024 (AIS, LACSEA, Sensors, QSM), paper JF3A.4
  14. [C2] F. Guzman, N. Diaz, E. Vera, Improved Compressive Temporal Imaging using a Shuffled Rolling Shutter, OSA Optical Sensors and Sensing Congress, virtual, 2021.
  15. [C3] J. Bacca, N. Diaz, H. Arguello, Compressive classification via deep learning using single-pixel measurement, 2020 Data Compression Conference, USA, 2020.
  16. [C4] N. Diaz, A. Basarab, J-Y Tourneret, H. Arguello, \textbf{Cardiac motion estimation} using convolutional sparse coding, 2019 27th (EUSIPCO), España, 2019.
  17. [C5] N. Diaz, A. Basarab, J-Y Tourneret, H. Arguello, Adaptive coded aperture design by motion estimation using convolutional sparse coding in compressive spectral video sensing, CAMSAP, France, 2019.
  18. [C6] N. Diaz, J. Bacca, H. Arguello, Gradient thresholding algorithm for adaptive colored coded aperture design in compressive spectral imaging, OSA Optical Sensors and Sensing Congress, USA, 2017.

Future Research Directions

My future research will pursue two strategic areas that build on my previous work while addressing critical challenges in computational imaging:


  1. Algorithms and Machine Learning for Computational Imaging: Novel algorithms that exploit enormous amounts of data have revolutionized the reconstruction in all areas of computer vision and computational imaging. Currently, it is crucial to design both the optical encoding and the reconstruction leveraging novel techniques such as End-to-End design, and physics-informed neural representation, in particular, implicit neural representation. Furthermore, a significant challenge in neural networks is explainability, as they are often considered black-box approaches. To address this challenge, novel methods such as deep image prior, unrolling neural networks promote explainability by converting the layers of the network into iterations of an optimization problem.


  2. Optimal Compressed Sensing: The acquisition of signals in high dimensions involves the design of sampling schemes that promote total uniform sampling (uniform packing) and on average uniform sampling (irregular packing). This area has profound applications in video, imaging, depth, phase retrieval, and Extended-depth-of-Field coding. The major challenge in this area is designing a sensing pattern that promotes the invertibility of the inverse problem, -- although sphere packing only guarantees optimal density for 2, 3, 8, and 24 dimensions. Moreover, wavefront coding design is used to solve specific tasks, such as extended-depth-of-field promoting uniformity in the modulation transfer function in multispectral images. Therefore, a novel computational imaging system demands clever encoding approaches that integrate novel optimal sensing approaches for different light sensing modalities.


Research and collaborations:

I also believe that interdisciplinary research is a crucial element and may provide exciting and important challenges. In this regard, I have been fortunate enough to have worked with scientists from several disciplines. I have worked with Dr. Henry Arguello, Dr. Hoover Rueda and Dr. Jorge Bacca three well-known researchers in compressive spectral imaging at Universidad Industrial de Santander, Department of Computer Science, Colombia, in projects concerning high dynamic range in compressive spectral imaging, adaptive compressive spectral imaging and video, and adaptive multisensor spectral image classification. During my postdoctoral stage, I have had the opportunity to design novel sensing methods, for instance, a shuffled rolling shutter camera, an optimal multispectral filter array, and single-mask sphere packing for ultrahigh-speed imaging with Dr. Esteban Vera at Pontificia Universidad Católica de Valparaíso, who is one of the leading researchers in area of computational imaging. During my postdoctoral stage, I spent six weeks visiting Dr. Pablo Meza at Universidad de la Frontera, School of Electrical Engineering, Chile. During this time, I learned about test-bed experiments for multispectral imaging and have come away with several applications in spectral video and sphere packing. Recently, I visited Dr. Jinyang Liang, who works at the Institut National de la Recherche Scientifique (INRS), at the Université du Québec, Canada and Miguel Marquez, postdoctoral fellow at the INRS. We designed a sensing scheme involving sphere packing and implicit neural representation for snapshot ultrahigh-speed imaging. Currently, I am working on a project with Dr. Karen Erguiazarian, a Professor of Signal Processing at the Department of Computing Sciences, Tampere University, Finland. And Dr. Samuel Pinilla, a researcher at Rutherford Appleton Laboratory, Science and Technology Facilities Council, UK. The project is about a lensless camera using a diffractive optical element to improve multispectral-extended-depth-of-field, I maintain active collaborations with both researchers. I have come to realize that research collaboration fosters novelty and innovation among researchers and scientists.