Project: Paint

The intention of this project was to faithfully recreate the characteristics of a camera lens "test scene," containing plentiful fine detail meant to increase complexity and visual noise.


One example of this concept in action is the DPReview (Digital Photography Review) test scene, which is used to compare the sharpness of lenses across brands. Their scene shares a similar construction and many of the same paper test patterns.

Textures for this project were created in Adobe Illustrator with noise added in Photoshop. In order to achieve the necessary detail, 8K textures are used on most objects throughout the scene.


Simple monochrome Gaussian noise was generated and overlaid in Photoshop to create the grainy finish present on most textures used in the scene. Although this gives the appearance of noise it also increases percieved detail, while adding a similar effect to dot matrix printing on low quality paper stock.


Some of the textures created can be seen on the left in reduced resolution. On the top left, a color wheel. On the top right, a moire circle, used in real-world tests to generate moire patterns. On the bottom left, a diagonal line, gradient, and text detail test. On the bottom right, one of the red paint tubes.

In terms of practical use, I used this scene to compare pixel filters (window functions) in contribution to the LuxCoreRender open source project.


Visually, the Sinc and Catmull-Rom functions include negative lobes, which create ringing artifacts around bright pixels, increasing percieved contrast and detail.


To the right are comparisons of three seperate areas of the scene, using four different pixel filters. The Sinc and Catmull-Rom functions were lost in the transition from LuxRender 1.X to 2.X, while the Blackmann-Harris and Gaussian functions remained.


As a result of the submission of this comparison, the Sinc and Catmull-Rom filters were added to the program, giving users further render options.


For further information on window functions and their usefulness in image reconstruction, I reccommend referencing this helpful chapter of PBRT.

LuxRender 1.3.1 & Blender 2.69