Radiance based method for accurate determination of volume scattering parameters using GPU-accelerated Monte Carlo

Abstract

Volume scattering is an important effect in different fields, ranging from biology to lighting. Models for volume scattering usually rely on parameters that are estimated with inverse methods that iteratively fit simulations to experimental data. To obtain accurate estimates for these parameters, the scattered intensity distribution can be used in such fitting methods. However, it has been shown that for samples with long optical path lengths this type of data may result in poor parameter estimates. In this work, an inverse procedure is proposed that fits to scattered radiance distributions. By taking advantage of current generation graphics processing units, the method implemented is sufficiently efficient to allow performing an in-depth simulation study on the difference between using radiance or intensity distributions to estimate the volume scattering parameters of samples. This work shows that for samples with moderate optical path lengths, the intensity distribution contains sufficient information to accurately estimate the volume scattering properties. However, for longer optical path lengths, the descriptive power of the intensity distribution is not enough and radiance distribution based methods, such as the inverse method proposed, are better suited.

Survey of Models for Acquiring the Optical Properties of Translucent Materials

“…Alternatively, we can use a more complete measurement of the scattered spectral radiance. Studies have shown that the fitting accuracy is usually better if a more complete measurement of the scattered light is used [LLA∗13, CHCM17]. So fitting to a measured BSSRDF, a measured spectral radiance distribution, or multiple camera images [GZB∗13, HLC18] will provide more accurate results than fitting to radiant intensity or only total transmission/reflection.…”

“…Objective functions based on calculating a norm between measurements and simulations are also found in practice, especially for inverse adding‐doubling methods [PvGW93, LMD∗14]. Other types of objective functions which focus on minimizing the differences between the shape of the simulated and measured scattered distributions have also been described, using the normalized cross‐correlation [LLA∗13, CCL∗16] or the cosine distance [CHCM17]. Similarly, when using measurements that can include a significant amount of variance or are very dense, it can be useful to define the objective function so that it minimizes the difference of aggregate data such as a histogram of the measured and simulated distributions [YX16].…”

“…Many different methods are available for solving the radiative transfer equation [vdH80, VHWF13], such as discrete ordinate or S n methods [FYH∗18] and spherical harmonics or P n methods [CSY∗15, DBL10]. Of interest are methods such as the Kubelka‐Munk model, widely adopted in papermaking and colour industries [All80], methods based on the adding‐doubling method, used in biomedical research [LCH∗18, Pra95] and lighting [LMD∗14, CCL∗16], and Monte Carlo methods [WJZ95, FB09, CHCM17, NGHJ18] which are ubiquitous in many fields.…”

“…One option is to evaluate BRDF [PH00] or BSSRDF [DLR∗09] for various optical properties. For lighting, important quantities such as the luminance or the radiant intensity distribution (beam pattern) are typically calculated using Monte Carlo ray tracing [CHCM17], while in biomedicine the same methods are used for calculating internal radiometric quantities such as the fluence distribution [FB09, WJZ95].…”

ValoMC: a Monte Carlo software and MATLAB toolbox for simulating light transport in biological tissue

“…Monte Carlo simulations have been used, for example, in studying photoacoustics [31,32], fluorescence tomography [33,34], time resolved diffuse spectroscopy [35,36], statistical models for photon transport [37], biophotonics of corals [38], dose optimization for photodynamic therapy [39] and laser treatment of port-wine stains [40]. Furthermore, the first studies in which Monte Carlo is utilised in the solution of the inverse problem of an optical imaging problem have been implemented [15,[41][42][43][44][45][46].…”