Abstract. We present an interactive system, which allows the user to produce drawings in a variety of traditional styles. It takes as input an image and performs semi-automatic tonal modeling. Our system shifts tedious technical aspects to the computer side, while providing the user with freedom on the creative and aesthetic side. The user has low-level control over stroke placement, and high-level control over the tone, smudging and amount of detail. The drawing is rendered in real-time. The basic component is a thresholding model of strokes that can simulate a large class of styles (e.g. pencil, charcoal, engraving).

It provides a controllable simulation of the variation of pencil pressure or stroke thickness
traditionally used in tonal modeling. We introduce a novel fast equilibration approach
for the resulting thresholding structure. The user can specify smudging and control the
amount of detail over each part of the drawing



Overview
This paper makes the following specific contributions:

Decoupling strokes and high-level attributes: The user independently controls the
placement of strokes, and higher-level properties such as tonal adjustment and precision.

Stroke model:We extend a thresholding technique to simulate a large class of styles such as pencil, charcoal, or engraving. Our model supports smudging, widely used in traditional drawing (usually by rubbing a finger or a piece of paper on the drawing).

Equilibration: Thresholding requires a matrix with a flat histogram, which can be obtained by equilibration. We present a new rapid probabilistic technique.

Interactive hardware implementation: Our drawing rendering algorithm is hardware-assisted and easy to implement.

Context: the art of drawing
Inspired by different authors [Edw99, Gai96, Hal64, Men76, Spe17], we distinguish the following aspects of drawing: Gestalt, perspective, outlines, tonal modeling, and control of detail. For each category, it is important to make a distinction between the aesthetic choice and the technical achievement of this intent. It is also crucial to under-stand in what these aspects can be similar and different in drawing and photography.





算法流程图：




Thresholding
In digital halftoning, tone is obtained by comparing at each pixel a threshold matrix T (x, y) against the target tone of the input image I (x, y). Depending on the result, the pixel is declared black or white

In our system, strokes are simulated using local threshold structures, which we call threshold textures. The initial principle of our rendering is simple. To render the drawing, we render all the strokes using their threshold texture, resulting in a global threshold matrix T (x, y), which is then compared to the target tones I (x, y) to obtain a black and white drawing. This is the key in decoupling stroke style from tonal value: the threshold texture defines the stroke style, while tone is obtained by thresholding with respect to the target tone




Stroke threshold texture
A threshold texture is essentially a threshold structure as described above. It has a flat histogram and is represented as a grey-level texture map. Threshold textures can be either acquired by scanning real strokes , or modeled directly. We scanned real strokes for pencil and charcoal, and modeled etching-style、strokes, and then applied a simple histogram flattening algorithm.



Stroke drawing interface
The user draws free-hand strokes using the mouse or a tablet. We use a B´ezier-fitting method enforcingG1 continuity for poly-B´ ezier curves[Sch88]. Strokes are organized into layers to facilitate editing or selection of groups of strokes. We propose rudimen-tary hatching tools based on the interpolation between one or two carrier strokes and random perturbations

Smudging
Traditional smudging softens a drawing by spreading graphite particles. In contrast to the technique by Sousa et al. , our implementation of smudging is purely phenomenological. It is an extension of the antialiasing technique