Algorithms for processing polarization-rich optical imaging data

Abstract

This work mainly focuses on signal processing issues related to continuous-wave, polarization-based direct imaging schemes. Here, we present a mathematical framework to analyze the performance of the Polarization Difference Imaging (PDI) and Polarization Modulation Imaging (PMI). We have considered three visualization parameters, namely, the polarization intensity (PI), Degree of Linear Polarization (DOLP) and polarization orientation (PO) for comparing these schemes. The first two parameters appear frequently in literature, possibly under different names. The last parameter, polarization orientation, has been introduced and elaborated in this thesis. We have also proposed some extensions/alternatives for the existing imaging and processing schemes and analyzed their advantages.

Theoretically and through Monte-Carlo simulations, we have studied the performance of these schemes under white and coloured noise conditions, concluding that, in general, the PMI gives better estimates of all the parameters. Experimental results corroborate our theoretical arguments. PMI is shown to give asymptotically efficient estimates of these parameters, whereas PDI is shown to give biased estimates of the first two and is also shown to be incapable of estimating PO. Moreover, it is shown that PDI is a particular case of PMI. The property of PDI, that it can yield estimates at lower variances has been recognized as its major strength. We have also shown that the three visualization parameters can be fused to form a colour image, giving a holistic view of the scene. We report the advantages of analyzing chunks of data and bootstrapped data under various circumstances.

Experiments were conducted to image objects through calibrated scattering media and natural media like mist, with successful results. Scattering media prepared with polystyrene microspheres of diameters 2.97m, 0.06m and 0.13m dispersed in water were used in our experiments. An intensified charge coupled device (CCD) camera was used to capture the images. Results showed that imaging could be performed beyond optical thickness of 40, for particles with 0.13m diameter. For larger particles, the depth to which we could image was much lesser. An experiment using an incoherent source yielded better results than with coherent sources, which we attribute to the speckle noise induced by coherent sources. We have suggested a harmonic based imaging scheme, which can perhaps be used when we have a mixture of scattering particles. We have also briefly touched upon the possible post processing that can be performed on the obtained results, and as an example, shown segmentation based on a PO imaging result.