|
This paper discusses on noise reduction in holographic data processing. Especially, for phase data in speckle metrology, related to a wrapped modulo 2π phase map, post processing is required to reduce random fluctuations such as the speckle phase decorrelation. The choice of the filtering algorithm is a challenge since a large variety of filtering schemes are available in literature. In addition, the metric for evaluating algorithms has to be discussed, especially regarding the need, or not, for a reference image. In order to assess the evaluations of the available de-noising strategies, we have constituted databases with simulated phase data. Then, 34 de-noising algorithms were chosen considering their efficiency in image processing and digital holography, as stationary wavelet transform based algorithm with Daubechies, symlets wavelet, curvelets, contourlets, BM3D (block matching 3D) algorithm (state of the art in the image processing), NL-means (non local) algorithms, 2D windowed Fourier transform and SPADEDH ; in addition, we consider classical methods such as Wiener, median and Gauss filtering, anisotropic filtering and Frost filter which was widely used in SAR (synthtic aperture radar) imaging. These schemes were evaluated through rankings provided by pertinent quality metrics. But, the main problem with quality metrics is that they are not self-sufficient in the sense that they require a noise-free reference phase fringe pattern in order to be computed. In practical situations, one only has a set of measurements to be processed, and no exact phase is available to evaluate the quality of processing. In order to bypass this limitation, a noise-free reference metric was developed, that means a metric which is directly capable of providing information on how efficient the filtering is, without the help of any reference measurement and by only considering the measured available phase data. So, we present and discuss on the rankings which were obtained, and in addition, we present a reference free metric adapted to phase data filtering in speckle metrology.
|
