Joseph Constantin

Image noise detection in global illumination methods based on FRVM

Global illumination methods based on stochastic techniques provide photo-realistic images. However, they are prone to stochastic perceptual noise that can be reduced by increasing the number of paths as proved by Monte Carlo theory. The problem of finding the required number of paths in order to ensure that human observers cannot perceive any noise is still open. Until now, we do not know precisely which features are considered by the human visual system (HVS) for the evaluation of the image quality. This paper proposes a relevant model to predict which image highlights perceptual noise by using fast relevance vector machine (FRVM). This model can then be used in any progressive stochastic global illumination method in order to find the visual convergence threshold of different parts of any image. A comparative study of this model with experimental psycho-visual scores demonstrates the good consistency between these scores and the model quality measures. The proposed model has also been compared with SVM model and gives competitive performances.

Control of a robot manipulator and pendubot system using artificial neural networks

The paper introduces artificial neural networks for the conventional control of robotic systems for better tracking performance. Different advanced dynamic control techniques are explained and a new second order recursive algorithm has been developed to tune the weights of the neural network. The problem of real-time control of a Pendubot system in difficult situations has been addressed. Examples, such as positioning and balancing structures, are presented and performances are compared to a conventional PD controller.

Perception of noise in global illumination algorithms based on spiking neural network

This paper proposes a reduced reference quality assessment model based on spiking neural network (SNN) in order to predict which image highlights perceptual noise in unbiased global illumination algorithms. These algorithms provide photo-realistic images by increasing the number of paths as proved by Monte Carlo theory. The objective is to find the number of paths that are required in order to ensure that most of the observers cannot perceive noise in any part of the image. A comparative study of this model with human psycho-visual scores demonstrates the good consistency between these scores and the learning model quality measures. The proposed model that uses a simple architecture composed only from two parallel spike pattern association neurons (SPANs) has been also compared with other learning model like SVM and gives satisfactory performance.

A generic Simulink based model of a wireless sensor node : Application to a medical healthcare system

Wearable low power sensor nodes enable medical facilities and personnel (doctors, nurses ...) to keep an eye on a persons vital signs without altering his daily routine. This technology is mainly limited by the amount of available energy. In this work, we aim to create a Matlab/Simulink based model that helps in the design of a wireless sensor node. We will not only be able to determine the power consumption of the node, but also to pick components based on their power-efficiency. This approach will allow the user to extract maximum performance from components already available on the market. Thus it will save the time needed to develop new ones. Examples with processors and wireless communication modules are presented. Future works will aim to upgrade the model in order to include several sensors operating within a network.

A new framework for analyzing the performance of the glucose-insulin system

The ability to develop a testing model for analyzing the performance of any embedded biomedical devices can be very useful in medical research. This paper begins by the glucose insulin system and creates a new framework in order to test the performance of all system components. The framework consists of simulating a mathematical model of human body in order to implement in a microcontroller, developing a control algorithm for the model and applying parametric models of activities to show how medical devices can be interconnected to form a physiological closed-loop system. Finally a new tester model with the ability to satisfy medical decision criteria has been created. Simulation results show the performance of the tester to validate glucose insulin system in order to avoid risk taking on patients.

An in-silico study for glucose-insulin system based on microcontroller using system simulator

There is a need for consistent approach to interpret and to optimize the testing of clinical medical equipment. We propose a model for global simulating of biomedical equipments (including human interaction models). This paper begins by an in-silico study for type 1 diabetes mellitus patients using a mathematical model, implementation of a simple control algorithm, than generating virtual patients in order to have multiple scenarios to analyze performance. Finally, we associate it with a control-variability grid analysis in order to do a step to construct a clinical trial simulator.

Visual Impact of Rendering on Image Quality

To generate photo-realistic images, modern renderers generally use stochastic algorithms such as the path-tracing algorithm. These algorithms can produce high-quality images but may require a long computation time. Therefore, rendering is generally stopped after a given amount of time and the output image may not be fully converged. In this case, the resulting variance can be seen as noise.

No-Reference Methods and Fuzzy Sets

This model can then be used in any progressive stochastic global illumination method in order to estimate the noise level of different parts of any image. A comparative study of this model with a simple test image demonstrates the good consistency between an added noise value and the results from the noise estimator.

Full-Reference Methods and Machine Learning

This chapter introduces the application of machine learning to Image Quality Assessment (IQA) in the case of computer-generated images. The classical learning machines, like SVMs, are quickly remained and RVMs are presented to deal with this particular IQA case (noise features learning). A recently performed psycho-visual experiment provides psycho-visual scores on some synthetic images (learning database), and comprehensive testing demonstrates the good consistency between these scores and the quality measures we obtain. The proposed measure has also been compared with close methods like RBFs, MLPs, and SVMs and gives satisfactory performance.

Reduced-Reference Methods

Reduced-reference image quality assessment needs no prior knowledge of reference image but only a minimal knowledge about processed images. A new reduced-reference image quality measure, based on SVMs and RVMs, using a supervised learning framework and synthetic images is proposed in this chapter. This new metric is compared with experimental psycho-visual data with success and shows that inductive learning is a good solution to deal with small sizes of the databases of computer-generated images. As reduced-reference techniques need only small size of labeled samples, thus the rapidity of the learning process is increased.