Warning: ini_set(): A session is active. You cannot change the session module's ini settings at this time in /usr/local/www/joomla/libraries/joomla/session/handler/joomla.php on line 48
Open Subjects

Open Subjects

PhD Subject: Human faces description using pixel-based and semantic-based analysis

Scope

This PhD is part of a project funded by the industry – Portuguese Mint and Official Printing Office – whose objective is to extract information from faces that can be used as security elements of ID cards, passports, airport gates, e-gates and other security documents or portals.

Objectives

The main objective of this PhD project is to study and develop new methods to extract information from human face images to build a unique descriptor of faces. Faces are used in several applications to identify humans and to rule the interaction between humans and machines.

One of the main problems in building huge databases of human faces is that the validation process can become intractable with medium-size or higher DBs. Therefore, the existence of description vectors of faces can be the best solution for real-time or almost real-time applications, since it represents a trade-off between accuracy on the validation process and the decision time, of key importance in industrial and human-machine systems.

Most of the existing solutions are pixel-based. Some paths to explore are based on semantic content of images which can extract useful information to describe the human face. The most simple examples include to describe the color of the skin, eyes, eyebrows, shape of the nose and other features using semantic interpretation and extraction. This data can accelerate the validation process, achieving considerable speed-ups in relation to pixel-based techniques. It if one of the main objectives of this work to explore this trade-off between pixel-based and semantic-based approaches.

In terms of utility of the approach, it will provide the necessary support for querying databases using natural language, or generate those descriptions about a person from a photo or a robot onboard camera and have it describe a person through speech.

Additionally, human faces express emotions all the time and facial expressions can be sufficiently strong to disturb the human validation process. It is also objective of this project to identify emotions in human faces and produce a descriptor agnostic of these emotions, so that the validation process can match two faces even if expressing different feelings or emotions.

Applications


Face identification for security documents
Human identification in e-gates of airports and other portals
Face identification for smartphones and mobile devices
Human-Robot Interaction

 

Advisors: Prof. Paulo Menezes and Prof. Nuno Gonçalves

VisualChain: Visual Inspection in a production chain

Objective: To develop a quality control system using computer vision to control the parameters of the plates of certain references of bicycle chains.

SRAMPORT is a company of the Chicago-based SRAM group and produces high-quality bicycle chains.
A chain is composed of outer plates, inner plates, rollers and shafts. The plates are produced in presses by a stamping process. There is now a manual control of the physical parameters of the plates produced.

The aim of this work is to develop a system that has the capability of measuring the physical characteristics of several plates dispersed on a base,

after the measurement the system will validate each of the sample plates or reject according to the parameters of the model and its tolerance .

In a second phase the system can undergo an evolution for continuous measurement and integration in the productive process.

Workplace: ISR-UC e SRAMPORT

Supervision: Prof. Paulo Menezes and Eng. Paulo Silva (SRAMPORT)

Spatial Augmented Reality for Serious Games

Most people have assisted to video-mapping shows where pictures and movies are typically projected using buildings structures as a screen.
We know that pointing a projector to a wall requires some adjustment of its orientation or the correction of the "trapezoidal distortion" (affine transformation) that appears when the optical axis is not perpendicular to the wall plane. When projecting over non-planar surfaces the distortion effects render the projected image hard to understand, depending on the relative position of the projector, the "screen" surface(s) and the viewer.

The objective of this work is to develop an interactive system composed of a Kinect device and a video projector that will be used to project over a table, a slanted wall or even the floor near a robot. Using new low consumption LED-based video projectors it is possible to embed it, together with a Kinect-based device on a mobile robot, that may then use this system to display information on the most appropriate nearby surface.
There are indeed several interesting possibilities for the use of this system like  (1) in making the projection over a set of objects and modify their appearance or create a scenario for a game based on the manipulation of these objects, or (2) in industrial contexts.

Although in this work we will aim at develop a game-based rehabilitation tool for stroke patients or elderly people, it will be closely related with an industrial application on the context of an ongoing research project.

Remarks:

This work will be benefit from a current ongoing project and from a collaboration with LAAS-CNRS a large research institution in Toulouse, France.
Depending on the pace of the evolution of the project, the student may be invited to do an internship at LAAS-CNRS co-advised by Prof. Frédéric Lerasle, where the working language can be either English or French depending on the student fluency on these languages.

Workplace: ISR-Coimbra / LAAS-CNRS in Toulouse, France

deepSTAIl: Style Transfer for Artificial Illustrations

Deep learning-based techniques are at the centre of the attentions of companies, researchers and even general public. The applications on the recognition of people, places and objects have demonstrated the power of these techniques showing unprecedented success rates.

One interesting application of neural networks is in what is called “style transfer”. This corresponds to get a picture or illustration and by using an exemplar from a specific painter the system produces a version of the input picture with close resemblance to the works of that painter.

Read more: deepSTAIl: Style Transfer for Artificial Illustrations

Haptic Interaction for Simulated Micromanipulation

Manipulation of microscopic objects is gaining relevance as the techniques evolve and bring the possibility of dealing of smaller and smaller objects. For example in biology the possibility of manipulating individual living cells, embryos and stem cell and generic therapies is of outmost importance. For example performing injection in cells is a task that requires extensive practice, in particular as it is performed at a level where the magnitude of the involved forces are very hard to measure, and only a 2D-like visualisation of the task is possible through a microscope. Given this, it is quite normal that in many cases the cells explode due to imprecise manipulation that in turn comes out from the lack of reliable feedback that may bring the operator to sense it properly. In this work we intend to develop a system for manipulating cells or other microscopic objects, based on the generation of haptic forces extracted from visual cues extracted from the manipulated objects’ deformations.

Read more: Haptic Interaction for Simulated Micromanipulation