ElectroScience: Paula Queluz, researcher of the PORTRAIT project

Paula Queluz, a professor at the Department of Electrical and Computer Engineering (DEEC) and a researcher at the Institute of Telecommunications (IT), shared some of her research in the field of image processing, including the PORTRAIT project – "Perceptual-aware geOmetrics sTurctuRe Adaptation of wIde-angle photTography," in which she is involved, as well as the study of omnidirectional (360°) images, in an interview with DEEC.

The PORTRAIT project, funded by a smartphone manufacturer, emerged from the PhD thesis of Falah Jabar, supervised by Paula Queluz and Professor João Ascenso. It proposed new methods for projecting omnidirectional (360°) images onto a plane, aiming to minimize the perception of distortions, particularly on smartphone screens. The project aims to adapt some of the methods developed in the thesis to images captured with a wide-angle view (wide-angle), using smartphones, as these images can also exhibit geometric distortions.

Since the projection of a sphere onto a plane inevitably induces geometric distortions, the proposed methods — based on the Pannini projection — optimize the parameters of this projection according to the content of the image, minimizing the perception of these deformations.

Paula Queluz, a professor at the DEEC and a researcher in the PORTRAIT project

But what is the Pannini projection, and what do these concepts mean in practice? Let’s start with photographs taken with a simple smartphone. As the wide-angle camera expands the field of view used to capture the image, distortions can occur, such as objects close to the camera appearing disproportionately larger than those in the background, objects at the edges of the image becoming distorted, and the central area of the image appearing farther away than it actually is. The Pannini projection aims to reduce these distortions.

The Pannini method is "a projection technique primarily used to generate panoramic images or images with wide angles (usually above 90 degrees), and it better preserves the proportions and perspective of the scene compared to linear projection (the most common in photography)." In the PORTRAIT project, the optimization of the Pannini projection uses distortion metrics applied both to the objects of interest and the background of the image, "reducing distortion both globally and locally without altering the spatial relationship between the various objects."

The Pannini projection starts from a spherical representation (i.e., 360°) of the scene, with the projection onto the plane carried out through two consecutive projections: from the sphere to an encompassing cylinder, and from the cylinder's surface to the plane.

Paula Queluz, a professor at the DEEC and a researcher in the PORTRAIT project

Regarding omnidirectional images, which were the focus of Falah Jabar’s thesis, the researcher mentions that there are applications in various contexts, such as in gaming and virtual reality, enabling the creation of increasingly immersive experiences. However, research in this area could also impact tourism and culture, creating a new way of perceiving different environments, such as museums, exhibitions, or other destinations, through virtual tours. In robotics and the development of autonomous vehicles, omnidirectional images are used to capture the surrounding environment, "allowing for detailed analysis for navigation and decision-making," she highlights. Lastly, she points out that there is also the potential to create a new type of object visualization in fields like architecture and design.

An example of a resource already used by many of us (…) is Google Street View, integrated into Google Maps, which allows users to virtually explore locations around the world, navigating through streets, cities, and environments.

Paula Queluz, a professor at the DEEC and a researcher in the PORTRAIT project

Paula Queluz also conducts research in the area of Radiance Fields (Neural Radiance Fields and Gaussian Splatting). Aiming to enhance the user’s immersive experience, these new forms of image representation enable the reconstruction of complex scenes from a set of 2D images, something that has revolutionized the representation of 3D images. Indeed, traditional omnidirectional images only allow the observer to explore the image through rotation or by moving along a predefined path, as seen in Google Street View, and the perspective of the captured images cannot be altered. By using Radiance Fields, it is possible to synthesize new images from different perspectives, providing continuous and free navigation within the 3D scene.

Based on neural networks, the NeRF model learns the radiance function of the scene, allowing it to estimate color and opacity at any point and direction within the three-dimensional space.

Paula Queluz, a professor at the DEEC and a researcher in the PORTRAIT project

When asked about how her interest in image processing arose, Paula Queluz highlights that it began during her master’s thesis in Electrical and Computer Engineering, but that she had always wanted to understand and be part of the technological revolution, especially in the fields of telecommunications and electronics. However, despite the growth of the tech sector in recent years, funding for projects remains limited, with job insecurity and a lack of stability in scientific research careers.

The researcher also points out the increasing demand for human resources in technological fields, in general, as these are the sectors "growing the most globally, yet facing a significant shortage of qualified professionals." In this context, she emphasizes the importance of including women, noting that many scientific fields, such as Electrical and Computer Engineering, have a significantly higher number of men. This may reflect the result of "historical, cultural, and social factors that, for decades, have influenced women's educational and professional choices."

Bringing more women into technological fields is essential because diversity is a factor of innovation. Teams made up of people with different perspectives, experiences, and ways of thinking have a greater ability to solve complex problems and create creative solutions.

Paula Queluz, a professor at the DEEC and a researcher in the PORTRAIT project

In recent years, several initiatives have been promoted to encourage women to pursue STEM (Science, Technology, Engineering, and Mathematics) careers, such as the group "Gender Balance @Técnico," which "aims to promote and enhance diversity at IST," she mentions. The inclusion of women in various fields is a growing trend, something that, in her view, "will not only help address this human resource need but also ensure that human potential is used more efficiently."

Tags: