Photometric stereo with non-parametric and spatially-varying reflectance

Photometric Stereo in the Wild

“…Representative work includes the example based methods [11,9] to transfer normals from reference or virtual objects to the target one; the glossy surface method using cast shadow for normal estimation [5]; bivariate approximation of the isotropic reflectance function to model spatially varying reflectance [3]; sub-pixel surface normal estimation method [24] using Expectation Maximization (EM); method to tackle the anisotropic reflectance effect [13]; the dense frame robust photometric stereo method [26] for casual data acquisition; colored lighting method [10] for dynamic objects; consensus method [12] to get rid of analytical models; and the self-calibration method [22] to automatically calibrate lighting information and scene radiance simultaneously.…”

A projective framework for radiometric image analysis

“…For comparison, the (generally distinct) iso-depth contours (recovered as in [1]) are shown in blue. The result of our procedure is shown on the right, and as a form of 'ground truth', we show the corresponding contours taken from the complete reconstructions obtained by Alldrin et al [2] 2 . Despite the differences between these two reconstruction procedures and 2 See additional results at http://vision.ucsd.edu/ nalldrin/research/ Figure 6.…”

“…This has been exploited, for example, for surface reconstruction using isotropy [5], bilateral symmetry [1,10], reciprocity [12], and half-vector symmetry [2]. The advantage of such symmetry-based approaches is that they avoid the use of low-parameter BRDF models (Lambertian, Ward, Cook-Torrance, etc.…”

BRDF Representation and Acquisition

“…Under certain assumptions, also setups used to acquire objects geometry through the classical photometric stereo technique [Woo80], where the point of view is kept constant between successive images while the direction of incident illumination varies, have been successfully used to recover BRDF and SVBRDF of non-lambertian surfaces [Geo03, GCHS05, HS05, CGS06,ZREB06,AZK08,HLHZ08]. Some of these techniques are limited to materials with a single specular lobe [GCHS05] due to the use of optimisation algorithms to recover the parameters for the Ward isotropic BRDF [War92] or require the acquisition of reference objects of known shape and with similar materials as the target [HS05]; to reduce the number of input pictures it has been assumed bivariate BRDFs [AZK08] or spatial coherence of reflectance, trading spatial for angular resolution [ZREB06]. The taxonomy of the acquisition setups, detailed in the following sections, is reported in Figure 9.…”