Perceptual evaluation of speckle noise reduction techniques for phase shifting holograms

The perceived image quality of a digital hologram of macroscopic objects is affected by its reduced depth of focus and by speckle noise due to coherent illumination. Several filtering techniques have been proposed for speckle noise reduction of digital holograms but there are scarce quality assessment studies regarding phase-shifting digital holograms. Typically, the performance of these filters on experimental holograms is assessed using no reference objective metrics. However, these metrics do not reflect the subjective visual quality perceived by a human observer. In this work, the performance of four speckle reduction algorithms, namely the non-local means, the Lee, the Frost and the block matching 3D filters, using five different parametrizations for each case, are subjectively compared. Due to a large amount of testing data, the subjective test was divided in two: 1) one phase where the effects of the parametrization of each method are evaluated, 2) a second phase where the best parametrization results are compared between them. The results are ranked with respect to the perceived image quality to obtain the Mean Opinion Scores for each filter/parameter combination. As in this case there is no reference image and double stimulus is desirable, the subjective evaluation was achieved using full paired comparison. The experiment indicates that BM3D and Lee are the preferred filters, and reveals a strong dependence of filter performance on hologram characteristics.