A Stochastic Film Grain Model for Resolution‐Independent Rendering

Abstract

The realistic synthesis and rendering of film grain is a crucial goal for many amateur and professional photographers and film-makers whose artistic works require the authentic feel of analog photography. The objective of this work is to propose an algorithm that reproduces the visual aspect of film grain texture on any digital image. Previous approaches to this problem either propose unrealistic models or simply blend scanned images of film grain with the digital image, in which case the result is inevitably limited by the quality and resolution of the initial scan. In this work, we introduce a stochastic model to approximate the physical reality of film grain, and propose a resolution-free rendering algorithm to simulate realistic film grain for any digital input image. By varying the parameters of this model, we can achieve a wide range of grain types. We demonstrate this by comparing our results with film grain examples from dedicated software, and show that our rendering results closely resemble these real film emulsions. In addition to realistic grain rendering, our resolution-free algorithm allows for any desired zoom factor, even down to the scale of the microscopic grains themselves.

Rendering Deformed Speckle Images with a Boolean Model

“…The random set P ⊂ R 2 follows a so-called Boolean model, a classical model of stochastic geometry [42,44,48] that has been applied in various situations such as, for instance, material analysis [45], distribution of trees in a forest [22], or film grain modeling in a resolution-free way [34] (see also [35] and the software implementation [36]). We shall come back to this latter paper in the discussion of the proposed algorithm.…”

“…It is known [42,48] that the proportion of 1-values (corresponding to a covering ratio of the random set P over the resulting image supported by Ω), is given by Table 1: Distribution of the random radii R ≥ 0 of the disks in the Boolean model. In the case of the log-normal law (used in the Boolean model of [34]), R is such that log(R) follows a Gaussian distribution of mean µ = log(µ 2 / µ 2 + σ 2 R ) and variance σ 2 = log(1 + σ 2 R /µ 2 ).…”

“…Note that, although the context is different, the authors of [34] also use Monte Carlo estimation to compute integrals for inhomogeneous Boolean models. These authors aim, however, at visual quality and not at accuracy, which permits them to drastically limit the computational burden.…”

“…The authors of [34] use the same random sequence (X m ) 1≤m≤N MC for any pixel x, which slightly improves the computation time. We do not adopt this simplification in order to avoid the results to be affected by a systematic bias.…”

Realistic Film Noise Generation Based on Experimental Noise Spectra

“…(1) The most straightforward way to incorporate noise is by adding a copy of stored film noise to the image. Although this deterministic approach is easy to implement, fast and may be reasonable for photos, it is particularly unfavourable to be applied to video sequences where each film frame would appear with exactly the same noise sample [6]. The noise is simply added in an uncorrelated manner to the signal (i.e.…”