1. Abstract
   1. what is assemblage
      1. kernel functions
      2. patterns for defining structured procedural textures
   2. contributions
      1. a new procedural random point distribution function
      2. a dynamic stochastic figure generation process
2. Introduction
   1. texture synthesis techniques
      1. by example[WLKT09]
      2. physical simulation[DRS08]
      3. procedure textures[EMP *98]
         1. advantages
            1. not depending on the surface size
            2. generate visual complexity with quasi-infinite variation at arbitrary definition while requiring only a marginal memory cost.
         2. disadvantages
            1. the creation of procedural textures is not a simple task
            2. the class of patterns that can be represented is limited
      4. our motivation
         1. propose a novel stochastic procedural pattern generation process
      5. core issue
         1. to be able to create infinite random variations of primitives and figures
      6. a hierarchical statistical shape model
         1. which allows us to represent shape variations, using statistical modes
      7. memory consumption & computational complexity
3. Related Works
   1. two main categories to do procedural texturing
      1. noise
         1. [LLC∗10]
      2. random point distributions
         1. represent a support function for
            1. sparse convolution noises [Lew89,vW91,LLDD09]
            2. cellular noises computed using n-th closest distances[Wor96]
            3. bombing patterns consisting of randomly "dropped" figures
   2. point distribution functions
      1. Jittering
      2. direct stochastic tiling.[LD05]
   3. two main categories to get the noise
      1. exploits the fact that large "texture shader" databases already exist [BD04, LLD12]
      2. derive procedural textures by using example images [WLKT09, LVLD10, GDS10, GLLD12].
4. Hierarchical statistical shape models for dynamic figure synthesis
   1. summerise
      1. fig2
   2. single-scale statistical figures
      1. discrete geometric elements -> continuous figure R(Φ)
      2. figure 2
      3. the goal
         1. to express the input set of sample figures in a uncorrelated orthogonal frame defined by matrix [Ψj]
   3. multi-scale figures
      1. figure5
5. Procedural object distributions
   1. using periodic rectilinear cell-based tessellations to define object distributions
      1. fig7
6. Designing and rendering procedural assemblage textures
   1. using the same mathematical formulation as for sparse convolution
      1. figure 9~11
7. Results
   1. a high-definition interactive texturing technique