This is a Timeline for the Newspaper company Berlingske, Berlingske have very very very long history, it funded in 260 years ago? even older than Hans Christian Andersen

This is just one of the small projects when I work in  in2media.dk

Very simple website, no complicate technology.

http://www.260.dk/

The 3-layer paper model

I implemented a new 3-layer model in my code, which I think is more relay on the fact than the author’s.

The first layer is advection layer, water and pigment flow on this layer in the rule of LBM and  also evaporat into air and saturate into fibre layer.

The second layer is fibre layer, the water and pigment saturated into this layer from advection layer by time step, and move to neighbor lattices  by the affection of capillarity, also sediment into the fixed layer.

The last layer is fixed layer, in this layer, water and pigment is not move anymore.

To make the organic pattern and arbitrary flow like the water flow on the rough surface, involve the restriction condition of paper is necessary, I implement the ‘fibre’ map from author’s thesis to restrict the flow speed and create a new noise map to simulate the paper saturability variance of each lattice.

  the saturability is created by perlin noise,http://en.wikipedia.org/wiki/Perlin_noise , perlin noise is a random noise pattern contain hierarchical details.

The original idea of this article is from Dr. Chu’s PhD thesis,（http://visgraph.cs.ust.hk/MoXi/ ）,In that project ,he proposed a novel approach for simulate brush and pigment interact with bibulous paper, especially calligraphy brush work with chinese “Xuan” paper. I tried to reconstruct the system during the vacation of Easter,  and come up some different idea for improve the simulation quality.

本文参考香港大学CHU SIU HANG的phd学位论文（http://visgraph.cs.ust.hk/MoXi/ ），并对作者的观点提出一些不同的看法和改进意见，探讨如何使用流体力学模型模拟颜料在可渗透材料上的物理运动，以实现尽可能逼真的画笔效果。

About the algorithm of flow simulation.

 In past decades,  several different numeric algorithm  of fluid simulation have been invented for computer graph implementation , the main idea of all these approach is subdivide the space into finite grids , each grid have it’s own  velocity and density,  the numeric value of velocity field and density field  evolved by time step ,in each time , the calculation is based a physical principle which come from discrete Navier – Stokes Equation  (figure 0  http://en.wikipedia.org/wiki/Navier-Stokes).

 I have tried 2 different algorithm, both of them are based on the  finite grids, I think finite grids is only way to simulation fluid on computer, the first one is called “Stable fluid simulation” from Jos Stam （http://www.dgp.toronto.edu/people/stam/reality/Research/pdf/GDC03.pdf ）he use 2 Poisson Equations to make mass conservation and momentum conservation , the other one is calle LBM (figure 2 http://en.wikipedia.org/wiki/Lattice_Boltzmann_Method ), which use only 1 Boltzmann Equation instead of 2 Poisson Equations.

关于 流体算法：

流体模拟物理原理来自于Navier-Stokes 方程（http://en.wikipedia.org/wiki/Navier-Stokes ）（图0），在视觉模拟领域，流体的模拟基本上采用半拉格朗日模型来实现，既通过有限元法将空间离散成nxm个网格，在每个时钟周期里，由网格来记载流体的速度场和密度场，而用粒子运动模型来模拟每一个网格内流体的运动。我尝试了2种流体算法，一种是论文中所提到的LBE（图1），另一种是多伦多大学Jos Stam发明的稳定流体算法。 （http://www.dgp.toronto.edu/people/stam/reality/Research/pdf/GDC03.pdf ）

I use the Jos Stam’s approach first, but I found  several problems hard to solve:

First ,Because of the algorithm based on very strict incompressible  rule, which will cause the density of all the lattices are equal together in each time step , but actually the water on the paper is not uniformly distributed, in practice, the area which just been inject water  (the drop droped on the paper or the brush touch the paper) contain very large amount of water (figure 2). the explanation from three-dimension  is the thickness  in that area is large , the explanation from  two-dimension is the density in that area is large, but the way of  Jos Stam’s approach force to uniform the density ,so it’s  can not reproduce this  phenomenon, unless use  three-dimensional simulation instead, then the gravity also need to be put in.

在使用 Jos Stam的算法时我发现了几个比较难以解决的问题：

因为 Jos的算法立足于十分严格的不可压缩条件，在2维的条件下，使用这个算法得到的解使每一个网格内流体的密度都是相等，而事实上水在纸张平面上的运动是有细微波浪的变化的，一颗水珠滴落到纸张上或者毛笔接触到纸张上的时候被注入水的网格先是含大量的水，（从3维的角度来解释就是网格内水的厚度比较高，从二维的角度解释是网格内水的数量大或者水的密度大），然后再逐渐散开（图2），如果用Jos的算法，这个过程2维方式无法模拟出来，必须用3维+重力的方式才能模拟。

 For explain this equation,  Jos’s approach discard the equibalance force of density variation, only consider equibalance force from saturation of  solute, it is equal the movement of fluid in the weightless condition, this is not realistic.

从方程的角度上来说，Jos的算法里丢弃了水趋往均势的运动量，只包含了水中溶解物趋往均势的运动量，相当于无重力环境下水在平面上的运动，或者是溶解物在液面上运动，这是不符合实际的。

 Second, because the density of lattices are restricted as constant, the other effects both absorbing  and evaporation will cause numeric change of density also can not be integrate into the algorithm .

 另外一点，因为网格含水量被严格限制为常量，对于外界能够干扰到水的密度数值变化的纸张的吸收、溶解、蒸发作用以及边界的进化等等都没有办法计算。

But  the LBE relaxed the condition of incompressible density, which means it’s allow the density change in lattices by time step,  and the difference of density in each neighbor lattice will affect the particle’s movement  direction, the explanation from  three-dimension is LBM allow waving , rippling and all kinds of numeric vary of water surface.(figure 3 and 5)

而LBE方程虽然也只是用2维来模拟水的运动，但LBE方程因为稀疏了不可压缩的条件，允许网格内含水量（方程表示为密度）的变化，在3维的情况下来看，就是允许了水的涟漪、波浪以及一系列水平面的数值变化，并且水平面的数值差异被方程处理为梯度和散度并最终形成作用力影响到水的运动方向（图3 和图5），其物理解释可以认为水在重力的作用下流动。

figure 4 demonstrat the incompressible algorithm cause the pigment unrealistic movement (no wet/dry edge,weightless )

I’ve read the paper about seam carving technology long time ago ,the arithmetic is invented in 2007, the main idea for this technology is about analysis the content of image and find the prior removable pixels when resize it, to make this work ,they find out a energy function to evaluate the importance of each pixel, and then find out the less important pixel path though top to bottom(or left to right), finally remove all of the pixels(or insert new pixels) align the pixel path , which been named as “seam”.

There are several ways to evaluate the importance of pixels , gradient evaluation is the easiest approach, based on information theory, if the color of pixel slice changed more acute, then there must be more information contain in that group of pixels, obviously it turns out the pixels should also be more important. 

But, the gradient map only demonstrate the importance of pixels relies on the vary of colors, that means the pixel of object edge marked as more important than the pixel inside the object shape , that is not true. Furthermore, some time the color vary of background frequent than color of foreground, but the fore ground should be the more important part.

So I think if we could integrate several different ways to evaluate the importance of pixels, that would get better result, in the thesis , the author also demonstrate the possibility of use human face detecting to find the face on the image, which must be the most important area, even let the user be able to draw some mark on the image mask, to tag the important area manually.

I’ve thinking about use some special way to analyses image, separate out the foreground and background  area, we can use morphologic erode arithmetic operators from the edge of image, step in to the center, the erode operator have some energy, each pixel meet the operator must add the energy of the operator , if meet the high gradient pixel in the image, then increase the operator’s energy ,  otherwise keep step forward.

I’ll keep working on this project this month, see if my idea works. 

About developing  in to flash, because flash doesn’t support change the size of Bitmap Data, I’ve forced to make a new class to store the binary data, which make the code more complex, I’ve spend many hours on figure out how flash ByteArray work with color and it’s read pointer stuff,  actionscript sometimes is really disgusting…

This is also a part of my graduation thesis project, for simulate the oil pigment flow on canvas, based on the Navie-Stokes equations  http://en.wikipedia.org/wiki/Navier-Stokes_equations

most of the ideas are got from

http://www.cs.berkeley.edu/b-cam/Papers/Bargteil-2006-ASL/

My master’s graduation thesis is about use computer to simulate the human painting process to draw a oil painting , I’v start some previously study of image analysis, this week I did this project for live face detection. I picked up several elegent approachs from several papers and integrate them together, get this great result.

the program now is still not stable enough, I’ll release the sourcecode when I fix the bugs.

This is funny. press mouse and draw the vine in the 3D space, and grow up some flower…

also with replay and record function.

Sometimes you will get bug when play the record , but I don’t care…I’m so tired, I have go to sleep now…

Source code always here when you open this link, and alwasy messy.

http://www.hxflyer.com/works/flowervine/RotationBelt.zip 

demo:

http://www.hxflyer.com/works/flowervine/demo/RotationBelt.html

here is the swf file for download:

http://www.hxflyer.com/works/flowervine/swf.zip

last year when I develop the panorama player in flashplayer 9, I have to build up extremelly complex 3D engine frist to support 3D matrix calculation and picture  3D distortion, for distort a picture in flash player 9, I have to subdivide the picture into a triangle matrix and render each triangle out,  the efficiency is terrible. now in flash player 10 ,to bulid up the 3d space is extremelly easy.

by the way ,the picture is shot by myself, in in2media office with my new fish eye lens!

demo:

http://www.hxflyer.com/works/panoflashplayer10/demo/pano.html

sourcecode in flex:

 http://www.hxflyer.com/works/panoflashplayer10/pano.zip

I’m working on flash player 10 3D programming these days, When I got an assignment to experiment google maps api, I tried combine it with flash player 10 3D engine, the result is looks far more fancy than I thought.

you can drag the red point to the map, move the map, rotate the 3D surface, even the usability sucks, but it’s cool! now I know why everybody like use the maps to make demo for their new equipment or any kinds of digital product showcase, the only reason is because demo made by map is fancy, even it doesn’t make any sense.

flash plyer 10 make 3D matrix calculation extremelly easy, this demo is only spend 3 hours work, if change to flash player 9,would be spend more than 1 day.

you should install flash player 10 to play this swf file.

open this link you can download my messy source code.

surce code:

http://www.hxflyer.com/works/googlemaps3d/GoogleMaps.zip

demo page:

http://www.hxflyer.com/works/googlemaps3d/GoogleMaps.html

  Last weekend I went back to china and bought a set of panorama device , a manfrotto 303 sph tripod head and a Sigma 8mm EX-DG fisheye lens,  which I can use this to take some virtual reality picture and maybe do some further research.

  I still need to buy a big tripod to support the head, Then I’m going to make a software can edit and distort the image.

the office:

the city hall of copenhagen: