Análise do comportamento de métodos de classificação de padrão de Speckle dinâmico

Abstract

The adoption of optical methods associated with computer vision can be an innovative and accessible alternative for the assessment of several types of materials. The laser speckle is one of them. When materials with moving particles are illuminated by a laser source they exhibit a dynamic speckle pattern called biospeckle. For the analysis of the phenomenon several sequential images must be acquired to portray its temporal behavior. The tools for the acquisition of speckle information are diverse due to the nonexistence of a single method for using on all kinds of materials. The purpose of this work was to better understand the Generalized Difference (GD) and Fujii methods, and to analyze them regarding concepts of computer vision. In the experiment, the drying process of a nail polish paint layer on a coin was evaluated. The relief difference on the coin provide for a lower concentration of paint on the image in relation to the background, what generates distinct patterns for each area. The studies led to the conclusion that the Fujii and Differences using the window (DW) based his estimates on the gradient, in other words, the extent of variation of a medium, in this case amplitude of the pixels. In the experiment, the GD and Fujii Equations were altered to improve and better understand the behavior of the processes, which, in some cases, presented new challenges for future works. Among the results, the deduction of a new GD algorithm using histograms lowered the GD execution time from 2 hours and 58 minutes to 21 minutes. This occurred because the complexity changed from O(n2) to O(n). The better understanding of the Fujii method set forth a proposal for a new Equation, with clearer results for this experiment, taking advantage of the improvement contained in the method.