Silvio de Barros Melo

Low Discrepancy Sequences Applied in Block Matching Motion Estimation Algorithms

This work presents a fast method for motion estimation by using low discrepancy sequences. The basic idea in this proposal is one based on the reduction of the computational effort involved in the matching of a given block with the reference block. Instead of using a metric that employs all corresponding pixels in both blocks, it uses one that selects a subset of pixels for which the coordinates are determined by the so called low discrepancy sequences. The proposed method is applied to the New Three Step Search (NTSS), typically used as a technique for motion estimation in video applications. The results show a reduction in computational complexity of about 80% with regard to the NTSS, with little degradation in the recovered frame. The block matching of this present method can couple nicely with virtually any motion estimation technique, requiring very little change.

A particle system approach to industrial tomographic reconstruction

This work introduces a novel tomographic reconstruction, using a very small number of projections, suitable for industrial applications. The particle system iterative reconstruction technique represents reconstructing material as a set of particles with certain physical features, subject to a group of forces, which can produce movement. The system constantly updates the net of particles by repositioning them in such a way as to approach the equilibrium. The elastic potential along a trajectory is a function of the difference between the attenuation coefficient in the current configuration and the corresponding input data. In this work, we present a simple analysis of the resulting reconstruction in a simulation with sinograms produced from some phantoms, including a comparison with some established iterative methods.

Quasi-Random Search algorithm for fast motion estimation

In this paper, we propose a new Motion Estimation algorithm based on low discrepancy sequences, essentially quasi-random sequences used as point sampling to compute multivariate integral approximations by methods such as Monte-Carlo. Our evaluation of the proposed method took into consideration PSNR mean values, computational effort and bit-rate. The results support that the proposed algorithm entitled Quasi-Random Search (QRS) is able to significantly reduce computational effort by 73,06% in average when compared to the UMHS algorithm while maintaining video quality and a slight increase in the bit-rate.

An Index of Susceptibility to Drought (ISD) for the Semiarid Brazilian Northeast

This work presents the development of a composite Index of Susceptibility to Drought (ISD) for semiarid Brazilian Northeast that considers climatology, physical properties, soil usage, social and economic aspects, the risk of harvest losses and the shortage of human and animal drinking water. The index started with the Index FUNCEME of drought Severity (IFS), developed by FUNCEME. Then, it evolved to use some tools proposed by the Joint Research Centre/Organization for Economic Co-operation and Development (JRC/OECD), as well as techniques of multiple imputation for missing data and data winsorization. The work was tested and validated with real data from Rio Grande do Norte State in three climatologic scenarios (dry, regular and rainy). A multivariate analysis test and a Monte Carlo simulation were also produced for a sensibility and strength analysis of the developed model. These analyses validated the composition model and the obtained results with real data. The ISD can be used as a tool to support decision makers in various government levels to help guide the actions for the drought-affected areas.

VDH-Grid Search Algorithm for Fast Motion Estimation

This work presents a fast block matching method for motion estimation algorithms by using low discrepancy sequences. The proposed technique (VDHS) was developed after analyzing the UMHS algorithm implemented within the H.264/AVC Reference Software. The optimizations are focused both in the reduction of the number of candidate blocks and the usage of pixel sub sampling based on VDH in order to accelerate SADs computation. Experimental results show that the proposed technique has a lower computational effort with an insignificant loss on PSNR and slight increment on the bit-rate.