Laplacian kernel splatting for efficient depth-of-field and motion blur synthesis or reconstruction

Abstract

Δ ∫ Δ Δ Input pixels Dense PSF Sparse PSF Laplacian domain Result a) b) c) d) 100 % 9,3 % Fig. 1. Computing motion blur and depth-of-field by applying a point spread function (PSF) to every pixel (a) is computationally costly. We suggest splating a pre-computed sparse approximation of the Laplacian of a PSF (b) to the Laplacian of an image (c) that under integration provides the same result (d). Note the circular bokeh combined with motion blur (1024×1024 pixels, 2 layers, 190 ms, Nvidia GTX 980Ti at .97 SSIM to a path-traced reference).Simulating combinations of depth-of-ield and motion blur is an important factor to cinematic quality in synthetic images but can take long to compute. Splatting the point-spread function (PSF) of every pixel is general and provides high quality, but requires prohibitive compute time. We accelerate this in two steps: In a pre-process we optimize for sparse representations of the Laplacian of all possible PSFs that we call spreadlets. At runtime, spreadlets can be splat eiciently to the Laplacian of an image. Integrating this image produces the inal result. Our approach scales faithfully to strong motion and large out-of-focus areas and compares favorably in speed and quality with of-line and interactive approaches. It is applicable to both synthesizing from pinhole as well as reconstructing from stochastic images, with or without layering.

Sample-based Monte Carlo denoising using a kernel-splatting network

“…Laplacian Kernel Splatting [Leimkühler et al 2018] accelerates large splats by exploiting sparsity in the Laplacian domain. It can be used for denoising, but in contrast to our model, it requires analytical expressions for the kernels.…”

Fast Monte Carlo Rendering via Multi-Resolution Sampling

“…We set the rendering resolution to 1920 × 1080 or 1080 × 1080 to cover the most content of scenes. For each image, we provide both the final rendered image and the render layers, which are essential for Monte Carlo rendering [2,3,9,23,24,28,31,32]. In total, each image has 33 rendering layers, including albedo, normals, depth, diffuse color, diffuse direct, diffuse indirect, glossy color and so on.…”

A Survey on Gradient‐Domain Rendering

“…The current gradient-domain techniques focus only on gradients, which is the first order image derivative. A recent work in image reconstruction [LSR18] shows that a higher order derivative (e.g. Laplacian) could be useful to reconstruct the image.…”