Posts Tagged ‘computer graphics’

Using feColorMatrix to dynamically recolor icons (part 2, two-color icons)

Previously, I looked at how to use feColorMatrix to dynamically change the color of single-color icons. In this post, we’ll look at using feColorMatrix, feBlend, and feComposite to implement an algorithm that allow for dynamically changing the the colors of an icon with 2 colors.

The Algorithm

With a single color, we could think of the process as being a single step: applying the color transformation matrix with the desired R, G, B values. For two colors, there are multiple steps and matrices involved, so it’s worth having a high-level overview and conceptual understanding of the algorithm before delving into the details.

  • The input icon will have 2 color, black and white; black areas will be changed to colorA and white areas will be changed to colorB
  • Add colorA to the source image (black areas will take on the new color, white areas will remain white), the result is imageA
  • Invert the source image, then add colorB to it (black areas will become white and remain white, white areas will become black and take on the new color), the result is imageB
  • Combine imageA and imageB, such that the alpha component from the source image is preserved, output the result
Dynamically recolor two-color icon

Note that from the above, we see the key operations that are needed:

  • Add
  • Invert
  • Combine

Another look at the color transformation matrix for applying a single-color

Note that the transformation matrix used previously for the single-color case, only preserves the alpha component from the input. The R, G, and B components are thrown away:

[[0, 0, 0, 0, R], [0, 0, 0, 0, G], [0, 0, 0, 0, B], [0, 0, 0, 1, 0]][(R_(src)), (G_(src)), (B_(src)), (A_(src))] = [(R), (G), (B), (A_(src))]

It doesn’t matter what color the input pixel is, applying the transformation matrix will result in those Rsrc Gsrc Bsrc input values being multiplied by zero, before the new/output R, G, B values are added in.

While this is fine for the single-color case, for the two-color algorithm to work, the distinction between the black areas and the white areas need to be kept intact, so we have to work with the Rsrc Gsrc Bsrc values from the input vector and preserve the distinction.

Making Rsrc Gsrc Bsrc part of the transformation

Modifying the transformation matrix to allow Rsrc Gsrc Bsrc to be part of the calculations requires the first 3 diagonal elements of the matrix to be non-zero.

The simplest case of this is the identity matrix:

[[1, 0, 0, 0, 0], [0, 1, 0, 0, 0], [0, 0, 1, 0, 0], [0, 0, 0, 1, 0]][(R_(src)), (G_(src)), (B_(src)), (A_(src))] = [(R_(src)), (G_(src)), (B_(src)), (A_(src))]

Let’s look at a few matrices that define transformations needed for the algorithm.

The add colorK matrix:

[[1, 0, 0, 0, R_(k)], [0, 1, 0, 0, G_(k)], [0, 0, 1, 0, B_(k)], [0, 0, 0, 1, 0]][(R_(src)), (G_(src)), (B_(src)), (A_(src))] = [(R_(src) + R_(k)), (G_(src) + G_(k)), (B_(src) + B_(k)), (A_(src))]

The invert matrix:

[[-1, 0, 0, 0, 1], [0, -1, 0, 0, 1], [0, 0, -1, 0, 1], [0, 0, 0, 1, 0]][(R_(src)), (G_(src)), (B_(src)), (A_(src))] = [(-R_(src) + 1), (-G_(src) + 1), (-B_(src) + 1), (A_(src))]

The above can be combined into a single transformation matrix, the invert & add colorK matrix:

[[-1, 0, 0, 0, 1 + R_(k)], [0, -1, 0, 0, 1 + G_(k)], [0, 0, -1, 0, 1 + B_(k)], [0, 0, 0, 1, 0]][(R_(src)), (G_(src)), (B_(src)), (A_(src))] = [(-R_(src) + 1 + R_(k)), (-G_(src) + 1 + G_(k)), (-B_(src) + 1 + B_(k)), (A_(src))]

Putting aside and referencing intermediate results

Note that the algorithm requires us to create 2 independent images (imageA and imageB) and then subsequently combine them.

This can be accomplished by utilizing the result attribute available on SVG filter primitive elements. By specifying a result identifier we are able to apply a transformation and put aside the resultant image.

Combining the intermediate results

Combining intermediate results/images can’t be done with feColorMatrix, it’s simply not an operation that can be constructed as a transformation. To handle this, SVG provides the feBlend element with a number of predefined operations available via the mode attribute. Based on the color transformations done to create imageA and imageB (mainly that the areas that are not of concern, the background, are set to white [1,1,1]), the multiply operation will work to combine the images (not perfectly, there’s one big problem with the alpha channel, but we’ll deal with that in a bit).

image multiply
<feBlend 
    color-interpolation-filters="linearRGB"
    in="imageA" 
    in2="imageB" 
    mode="multiply" 
    result="output" />        

The alpha channel problem

While feBlend seem to accomplish what’s needed, for anything other than a white background you’ll notice a white-ist outline around elements in the image, as you can see below.

alpha channel problems with feBlend

The problem is that for what we’re trying to accomplish we just want to simply preserve the alpha channel, not blend it in any way. However, all feBlend operations will perform a blend of the alpha channel. From the SVG spec:

image multiply

So there’s no way we can really work with feBlend to get a solution for the alpha channel, but we do have an idea of what the solution is: copy the alpha channel from the source image.

Fixing the alpha channel

Fixing the alpha channel will involve 2 steps:

  • For the image outputted by feBlend, get the alpha value to be 1 for every pixel
    (the reason this step will become apparent once we look at how feComposite has to be used)
  • Use feComposite to construct an image with the alpha values from the source image and the R,G,B values from the image outputted by feBlend

The first step is simple. We just need a slight modification to the 2 color transformation matrices used, such that one or both set the alpha channel to 1 for every pixel (note from the alpha blending equation, this will effectively set the alpha value to 1 for every pixel in the output). The modified matrices are shown below and it’s a good point to start showing some code.

The full-alpha add colorK matrix:

[[1, 0, 0, 0, R_(k)], [0, 1, 0, 0, G_(k)], [0, 0, 1, 0, B_(k)], [0, 0, 0, 0, 1]][(R_(src)), (G_(src)), (B_(src)), (A_(src))] = [(R_(src) + R_(k)), (G_(src) + G_(k)), (B_(src) + B_(k)), (1) ]

<feColorMatrix in="SourceGraphic" type="matrix" result="imageA"
    values="1 0 0 0 0
            0 1 0 0 0.68235294117
            0 0 1 0 0.93725490196
            0 0 0 0 1" /> 

The full-alpha invert & add colorK matrix:

[[-1, 0, 0, 0, 1 + R_(k)], [0, -1, 0, 0, 1 + G_(k)], [0, 0, -1, 0, 1 + B_(k)], [0, 0, 0, 0, 1]][(R_(src)), (G_(src)), (B_(src)), (A_(src))] = [(-R_(src) + 1 + R_(k)), (-G_(src) + 1 + G_(k)), (-B_(src) + 1 + B_(k)), (1)]

<feColorMatrix in="SourceGraphic" type="matrix" result="imageA"
    values="-1 0 0 0 1.0784313725
            0 -1 0 0 1.7058823529
            0 0 -1 0 1.2431372549
            0 0 0 0 1" />

For the next and final step, we need to take a look at feComposite.

The feComposite filter primitive allows for more fine-grained operations on pixels using the supported “arithmetic” operation:

feComposite arithmetic operation

As this operation is done on every channel, including the alpha channel, we have way to control what happens to the alpha pixels in a composition of 2 images.

We’re going to make use of this by:

  • Using an feColorMatrix, where the input in the source image, and transforming such that R,G,B is white [1,1,1] and the alpha remains unchanged
  • Using feComposite to do a simple arithmetic multiply (k1=1, k2=0, k3=0, k4=0), between the image constructed above (where R,G,B is [1,1,1] and alpha is the alpha from the source image) and the output from feBlend (where the R,G,B = the values we want for the output and alpha = 1)

Effectively, the source alpha is multiplied by 1 (as the image produced from the feBlend operation has the alpha set to 1 for all pixels) and the R,G,B values from the feBlend output are multiplied by 1 (as the constructed image, in the first step above, sets R,G,B to 1 for every pixel).

<!-- Get and use alpha from source image -->
<feColorMatrix in="SourceGraphic" type="matrix" result="whiteAlpha"
    values="0 0 0 0 1
            0 0 0 0 1
            0 0 0 0 1
            0 0 0 1 0" />     

<feComposite in="whiteAlpha" in2="outputFullAlpha" operator="arithmetic" k1="1" k2="0" k3="0" k4="0" />

Pulling everything together

We now have all the pieces for the filter and here’s what the code looks like:

<svg style="width:0; height:0; margin:0; padding:0; border:none;">
    <filter color-interpolation-filters="sRGB" id="colorTransformFilter">

        <feColorMatrix in="SourceGraphic" type="matrix" result="imageA"
            values="1 0 0 0 0
                    0 1 0 0 0.68235294117
                    0 0 1 0 0.93725490196
                    0 0 0 0 1" /> 

        <feColorMatrix in="SourceGraphic" type="matrix" result="imageB"
            values="-1 0 0 0 1.0784313725
                    0 -1 0 0 1.7058823529
                    0 0 -1 0 1.2431372549
                    0 0 0 0 1" />              
                    
        <feBlend 
            color-interpolation-filters="linearRGB"
            in="imageA" 
            in2="imageB" 
            mode="multiply" 
            result="outputFullAlpha" />        

        <!-- Get and use alpha from source image -->
        <feColorMatrix in="SourceGraphic" type="matrix" result="whiteAlpha"
            values="0 0 0 0 1
                    0 0 0 0 1
                    0 0 0 0 1
                    0 0 0 1 0" />     

        <feComposite in="whiteAlpha" in2="outputFullAlpha" operator="arithmetic" k1="1" k2="0" k3="0" k4="0" />

    </filter>
</svg>

The demo below uses the filter, along with a bit of Javascript to cycle and update the input colors (i.e. dynamically updating the values plugged into the first 2 feColorMatrix elements):

Limitations

We have the same limitations I mentioned in part 1:

  • For icons applied as background-images, the CSS filter property isn’t ideal. CSS filter will effect not only the element it’s applied to, but all child elements as well
  • As is the case with mixing the CSS filter property and the SVG filter element, effects governed by the CSS transition property won’t work

In addition, because of how the alpha channel in treated in regards to feBlend (setting all pixels to have alpha=1), you more than likely won’t get good results if the icon has different-colored adjoining or overlapping elements, as you won’t get a smooth transition at the edges/boundaries.

Using feColorMatrix to dynamically recolor icons (part 1, single-color icons)

I’ve been experimenting with using feColorMatrix as an elegant way to dynamically color/re-color SVG icons. Here I’ll look at working with single-color icons.

Working directly with the SVG markup

Changing the stroke and/or fill colors of the SVG elements directly can be a good solution in many cases, but it requires:

  • Placing the SVG markup into the document to query and modify the appropriate elements when a color update is needed (note that this option isn’t viable if you need to place the icon in an <img> tag or it needs to be placed as a background-image on an element, as you can’t reference the SVG element in such cases)
  • Treating the SVG markup as a templated, Javascript, string to make a data-URI, and re-making it when a color update is needed

By using the color transformation matrix provided by feColorMatrix these restrictions go away and we also get back the flexibility of using external files.

Icon color

Keep in mind, we’re only dealing with single-color icons. What color is used doesn’t technically matter, but black is a nice basis and in an actual project, black is beneficial, as you’re able to open your the icon files in an editor or browser and actually see it.

black colored icon

A black pixel within the icon can then be represented by the following vector:

black pixel as column vector

Note that the alpha component may vary due to antialiasing (to smooth out edges) or some translucency within the icon.

The color transformation matrix

feColorMatrix allows you to define a 5×4 transformation matrix. There’s a lot you can do with that, but note that the last column of the matrix is essentially an additive component for each channel (see matrix-vector multiplication), so in that column we enter the desired R, G, B values from top to bottom, which will be added to the zeros in the input vector. Next, we want to preserve the alpha component from the input vector, so the fourth column of the matrix becomes [0, 0, 0, 1]T and the fourth row of the last column is zero, as we don’t want to add anything to the alpha component.

color transformation matrix, [[0, 0, 0, 0, R], [0, 0, 0, 0, G], [0, 0, 0, 0, B], [0, 0, 0, 1, 0]][(0), (0), (0), (A_(src))] = [(R), (G), (B), (A_(src))]

The matrix-vector multiplication gives a new vector (that defines the output pixel) with the entered R, G, B values and the alpha value from the source pixel.

Representing the matrix within an feColorMatrix element is straightforward…

<feColorMatrix in="SourceGraphic" type="matrix"
values="0 0 0 0 R
0 0 0 0 G
0 0 0 0 B
0 0 0 1 0"
/>

… just plug in values for R, G, B.

Applying the color transformation

The color transformation matrix can be applied to an element by wrapping it in an SVG filter element and referencing the filter via the CSS filter property.

With Javascript, the values attribute of the feColorMatrix element can be updated dynamically. The color change will, in turn, be reflected in any elements referencing the SVG filter.

<!DOCTYPE html>
<html>
<body>
<!--
The values of the color matrix defines the color transformation what will be applied.
Here we just setup the elements and define an identity matrix. We'll modify the matrix via Javascript code
to define an actual color transformation.
-->
<svg style="width:0; height:0; margin:0; padding:0; border:none;">
<filter
color-interpolation-filters="sRGB" id="colorTransformFilter">
<feColorMatrix
in="SourceGraphic" type="matrix"
values="1 0 0 0 0
0 1 0 0 0
0 0 1 0 0
0 0 0 1 0"
/>

</filter>
</svg>

<!--
Element with an SVG icon that we want to colorize
Note: that the color transformation is applied to everything not only to the background, but everything
within the element as well.

Typical solution to to isolate background stuff to it's own div and use another div for contents

-->
<div id="logo-colored"
style="
width:300px;
height:300px;
background-color: transparent;
background-image: url(logo.svg);
background-position: center center;
background-repeat: no-repeat;
background-size: cover;
filter:url(#colorTransformFilter);"
>

<p
style="color:#fff">Testing 1 2 3...</p>
</div>

<script
type="text/javascript">
/**
* A little helper function to update the color transformation matrix
*
* @param {Number} _r
* @param {Number} _g
* @param {Number} _b
*/
const setPrimaryColor = function(_r, _g, _b) {

const rScaled = _r / 255.0;
const gScaled = _g / 255.0;
const bScaled = _b / 255.0;

const feColorMatrixElem = document.getElementById('colorTransformFilter').getElementsByTagName('feColorMatrix')[0];
feColorMatrixElem.setAttribute(
`values`,
`0 0 0 0 ${rScaled}
0 0 0 0 ${gScaled}
0 0 0 0 ${bScaled}
0 0 0 1 0`
);
};

// Set/update color transformation matrix
setPrimaryColor(129, 0, 0);
</script>

</body>
</html>

The code will take a black-colored icon and re-color it to [129, 0, 0], as seen below:

Limitations

This technique provides a lot of flexibility, but it’s not without it’s limits.

  • For icons applied as background-images, the CSS filter property isn’t ideal. CSS filter will effect not only the element it’s applied to, but all child elements as well. Note the “Testing 1 2 3…” paragraph is re-colored in the demo above.
  • As is the case with mixing the CSS filter property and the SVG filter element, effects governed by the CSS transition property won’t work.

Similar techniques

  • For shifting between colors, the hue-rotate() CSS filter function can be a solution. However, in practice, I don’t find this intuitive and color changes are rarely just hue rotations.
  • A more limited case, transitioning a colored icon to white, can be achieved with 2 CSS filter functions, brightness(0) and invert(100%).
  • You can do crazier things by trying to compute and fit a solution to the hue-rotation, saturation, sepia, and invert filter functions; however this is both complex to grasp and produces inexact/approximate color matches.
  • An SVG filter using feComponentTransfer should work, but I don’t find it as intuitive to work with.

Resources

If you want to interactively play around with feColorMatrix, check out SVG Color Filter Playground.

Encoding MP4s in the browser

Is this possible?

Given that it’s relatively easy to access a camera and capture frames within a browser, I began wondering it there was a way to encode frames and create a video within the browser as well. I can see a few benefits to doing this, perhaps the biggest being that you can move some very computationally expensive work to front-end, avoiding the need to setup and scale a process to do this server-side.

I searched a bit and first came across Whammy as a potential solution, which take a number of WebP images and creates a WebM video. However, only Chrome will let you easily get data from a canvas element as image/webp (see HTMLCanvasElement.toDataURL docs). The non-easy way is to read the pixel values from the canvas element and encode them as WebP. However, I also couldn’t find any existing JS modules that did this (only a few NodeJS wrappers for the server-side cwebp application) and writing an encoder was a much bigger project that I didn’t want to undertake.

The other option I came across, and used, was ffmpeg.js. This is a really interesting project, it’s a port of ffmpeg via Emscripten to JS code which can be run in browsers that support WebAssembly.

Grabbing frames

My previous post on real-time image processing covers how to setup the video stream, take a snapshot, and render it to a canvas element. To work with ffmpeg.js, you’ll additionally need the frame’s pixels from the canvas element as a JPEG image, represented as bytes in a Uint8Array. This can be done as follows:

var dataUri = canvas.toDataURL("image/jpeg", 1);
var jpegBytes = convertDataURIToBinary(dataUri);

convertDataURIToBinary() is the following method, which will take the data-uri representation of the JPEG data and transform it into a Uint8Array:

function convertDataURIToBinary(dataURI) {
var base64 = dataURI.substring(23);
var raw = window.atob(base64);
var rawLength = raw.length;

var array = new Uint8Array(new ArrayBuffer(rawLength));
for (i = 0; i < rawLength; i++) {
array[i] = raw.charCodeAt(i);
}
return array;
};

FYI, this is just a slight modification of a method I found in this gist.

Note that I did not use PNG images due to an issue in the current version of ffmpeg.js (v3.1.9001).

Working with ffmpeg.js

ffmpeg.js comes with a Web Worker wrapper (ffmpeg-worker-mp4.js), which is really nice as you can run “ffmpeg –whatever” by just posting a message to the worker, and get the status/result via messages posted backed to the caller via Worker.onmessage.

var worker = new Worker("node_modules/ffmpeg.js/ffmpeg-worker-mp4.js");
worker.onmessage =
function (e) {
var msg = e.data;

switch (msg.type) {
case "ready":
console.log(
'mp4 worker ready');
break;
case "stdout":
console.log(msg.data);
break;
case "stderr":
console.log(msg.data);
break;

case "done":
var blob = new Blob([msg.data.MEMFS[0].data], {
type:
"video/mp4"
});

// ...
break;

case "exit":
console.log(
"Process exited with code " + msg.data);
break;
}
};

Input and output of files is handled by MEMFS (one of the virtual file systems supported by Emscripten). On the “done” message from ffmpeg.js, you can access the output files via the msg.data.MEMFS array (shown above). Input files are specified via an array in the call to worker.postMessage (shown below).

worker.postMessage(
{
type:
"run",
TOTAL_MEMORY: 268435456,
MEMFS: [
{
name:
"input.jpeg",
data: jpegBytes
}
],
arguments: [
"-r", "60", "-i", "input.jpeg", "-aspect", "16/9", "-c:v", "libx264", "-crf", "1", "-vf", "scale=1280:720", "-pix_fmt", "yuv420p", "-vb", "20M", "out.mp4"]
}
);

Limitations

With a bunch of frames captured from the video stream, I began pushing them through ffmpeg.js to encode a H.264 MP4 at 720p, and things started to blow up. There were 2 big issues:

  • Video encoding is no doubt a memory intensive operation, but even for a few dozen frames I could never give ffmpeg.js enough. I tried playing around with the TOTAL_MEMORY prop in the worker.postMessage call, but if it’s too low ffmpeg.js runs out of memory and if it’s too high ffmpeg.js fails to allocate memory.
  • Browser support issues. Support issues aren’t surprising here given that WebAssembly is still experimental. The short of it is: things work well in Chrome and Firefox on desktop. For Edge or Chrome on a mobile device, things work for a while before the browser crashes. For iOS there is no support.

Hacking something together

The browser issues were intractable, but support on Chrome and Firefox was good enough more me, and I felt I could work around the memory limitations. Lowering the memory footprint was a matter of either:

  • Reducing the resolution of each frame
  • Reducing the number of frames

I opted for the latter. My plan was to make a small web application to allow someone to easily capture and create time-lapse videos, so I had ffmpeg.js encode just 1 frame to a H.264 MP4, send that MP4 to the server, and then use ffmpeg’s concat demuxer on the server-side to progressively concatenate each individual MP4 file into a single MP4 video. What this enables is for the more costly encoding work to the done client-side and the cheaper concatenation work to be done server-side.

Time Stream was the end result.

Here’s a time-lapse video created using an old laptop and a webcam taped onto my balcony:

This sort of hybrid solution works well. Overall, I’m happy with the results, but would love the eliminate the server-side ffmpeg dependency outright, so I’m looking forward to seeing Web Assembly support expand and improve across browsers.

More generally, it’s interesting to push these types of computationally intensive tasks to the front-end, and I think it presents some interesting possibilities for architecting and scaling web applications.

Brute-force convex hull construction

I’ve been experimenting a bit with convex hull constructions and below I’ll explain how to do a brute-force construction of a hull.

It’s worth noting up-front that the brute-force method is slow, O(n3) worst case complexity. So why bother? I think there are a few compelling reasons:

  • The brute-force method expresses the fundamental solution, which gives you the basic building blocks and understanding to approach more complex solutions
  • It’s faster to implement
  • It’s still a viable solution when n is small, and n is usually small.

What is a convex hull?

You can find a formal definition on Wikipedia. Informally, and specific to computational geometry, the convex hull is a convex polygon in which all points are either vertices of said polygon or enclosed within the polygon.

Brute-force construction

  • Iterate over every pair of points (p,q)
  • If all the other points are to the right (or left, depending on implementation) of the line formed by (p,q), the segment (p,q) is part of our result set (i.e. it’s part of the convex hull)

Here’s the top-level code that handles the iteration and construction of resulting line segments:

/**
* Compute convex hull
*/
var computeConvexHull = function() {
console.log(
"--- ");

for(var i=0; i<points.length; i++) {
for(var j=0; j<points.length; j++) {
if(i === j) {
continue;
}

var ptI = points[i];
var ptJ = points[j];

// Do all other points lie within the half-plane to the right
var allPointsOnTheRight = true;
for(var k=0; k<points.length; k++) {
if(k === i || k === j) {
continue;
}

var d = whichSideOfLine(ptI, ptJ, points[k]);
if(d < 0) {
allPointsOnTheRight =
false;
break;
}
}

if(allPointsOnTheRight) {
console.log(
"segment " + i + " to " + j);
var pointAScreen = cartToScreen(ptI, getDocumentWidth(), getDocumentHeight());
var pointBScreen = cartToScreen(ptJ, getDocumentWidth(), getDocumentHeight());
drawLineSegment(pointAScreen, pointBScreen);
}

}
}
};

The “secret sauce” is the whichSideOfLine() method:

/**
* Determine which side of a line a given point is on
*/
var whichSideOfLine = function(lineEndptA, lineEndptB, ptSubject) {
return (ptSubject.x - lineEndptA.x) * (lineEndptB.y - lineEndptA.y) - (ptSubject.y - lineEndptA.y) * (lineEndptB.x - lineEndptA.x);
};

This is a bit of linear algebra derived from the general equation for a line.

The result represents the side of a line a point is one, based on the sign of the result. We can check if the point is on the left or on the right, it doesn’t matter as long as there is consistency and the same check is done for all points.

How it looks

I made a few diagrams to show the first few steps in the algorithm, as segments constituting the convex hull are found. The shaded area represents our success case, where all other points are to the right of the line formed by the points under consideration. Not shown are the failure cases (i.e. one or more points are on the left of the line formed by the points under consideration).

convex hull construction, brute force, step 1

convex hull construction, brute force, step 1

convex hull construction, brute force, step 1

Code and Demo

You can play around with constructing a hull below by double-clicking to add vertices.

You can find the code on GitHub.

Post-process shaders in glfx

Pushed an update to glfx to allow for post-process shading. When a post-process shader is defined, the scene is rendered to a screen-space quad (the size of the viewport), and that quad is then rendered to the viewport with the post-process shader applied.

The shader is loaded (asynchronously) like any other:

glfx.shaders.load('screenspace.fs', "frag-shader-screenspace", glfx.gl.FRAGMENT_SHADER);

Once loaded, we create the shader program, and get locations for whatever variables are used. The vertex shader isn’t anything special, it just transforms a vertex by the model-view and projection matrices, and passes along the texture coordinates.

glfx.whenAssetsLoaded(function() {

var postProcessShaderProgram = glfx.shaders.createProgram([glfx.shaders.buffer['vert-shader-basic'], glfx.shaders.buffer['frag-shader-screenspace']],
function(_shprog) {

// Setup variables for shader program
_shprog.vertexPositionAttribute = glfx.gl.getAttribLocation(_shprog, "aVertexPosition");
_shprog.pMatrixUniform = glfx.gl.getUniformLocation(_shprog,
"uPMatrix");
_shprog.mvMatrixUniform = glfx.gl.getUniformLocation(_shprog,
"uMVMatrix");
_shprog.textureCoordAttribute = glfx.gl.getAttribLocation(_shprog,
"aTextureCoord");

_shprog.uPeriod = glfx.gl.getUniformLocation(_shprog,
"uPeriod");
_shprog.uSceneWidth = glfx.gl.getUniformLocation(_shprog,
"uSceneWidth");
_shprog.uSceneHeight = glfx.gl.getUniformLocation(_shprog,
"uSceneHeight");

glfx.gl.enableVertexAttribArray(_shprog.vertexPositionAttribute);
glfx.gl.enableVertexAttribArray(_shprog.textureCoordAttribute);

});

...

We then tell glfx to apply our post-process shader program:

glfx.scene.setPostProcessShaderProgram(postProcessShaderProgram);

This call will result in different rendering path, which renders the scene to a texture, applies that texture to a screen-space quad, and renders the quad with the post-process shader.

Here is the shader for screenspace.fs, used in the demo shown above:

precision mediump float;

uniform float uPeriod;
uniform float uSceneWidth;
uniform float uSceneHeight;
uniform sampler2D uSampler;        
varying vec2 vTextureCoord;

void main(void) {

vec4 sum = vec4( 0. );
float blurSampleOffsetScale = 2.8;
float px = (1.0 / uSceneWidth) * blurSampleOffsetScale;
float py = (1.0 / uSceneHeight) * blurSampleOffsetScale;

vec4 src = texture2D( uSampler, ( vTextureCoord ) );

sum += texture2D( uSampler, ( vTextureCoord + vec2(-px, 0) ) );
sum += texture2D( uSampler, ( vTextureCoord + vec2(-px, -py) ) );
sum += texture2D( uSampler, ( vTextureCoord + vec2(0, -py) ) );
sum += texture2D( uSampler, ( vTextureCoord + vec2(px, -py) ) );
sum += texture2D( uSampler, ( vTextureCoord + vec2(px, 0) ) );
sum += texture2D( uSampler, ( vTextureCoord + vec2(px, py) ) );
sum += texture2D( uSampler, ( vTextureCoord + vec2(0, py) ) );
sum += texture2D( uSampler, ( vTextureCoord + vec2(-px, py) ) );
sum += src;

sum = sum / 9.0;

gl_FragColor = src + (sum * 2.5 * uPeriod);

}

Note that it requires a few uniforms to be supplied to it, we use the glfx.scene.onPostProcessPreDraw() callback to setup the variables (before the post-processed scene is drawn):

var timeAcc = 0;
glfx.scene.onPostProcessPreDraw =
function(tdelta) {

timeAcc += tdelta;
var timeScaled = timeAcc * 0.00107;

if(timeScaled > 2.0*Math.PI) {
timeScaled = 0;
timeAcc = 0;
}

var period = Math.cos(timeScaled);
glfx.gl.uniform1f(postProcessShaderProgram.uPeriod, period + 1.0);

glfx.gl.uniform1f(postProcessShaderProgram.uSceneWidth, glfx.gl.viewportWidth);
glfx.gl.uniform1f(postProcessShaderProgram.uSceneHeight, glfx.gl.viewportHeight);
};

What we’re doing is using the scene rendering time deltas to generate a periodic/sinusoidal wave. This results in the pulsing brightness/fading effect of the scene. The brightness effect itself is done by adding the source pixel to a blurred + brightened version of itself. The blurring allows for the soft fade in and fade out.

GLSL variable qualifiers

I’ve been playing around with WebGL shader code recently and found this bit on variable prefixes helpful, particularly in the explanation of the variable qualifiers:

  • Attribute: data provided by buffers
  • Uniform: inputs to the shaders
  • Varying: values passed from a vertex shader to a fragment shader and interpolated (or varied) between the vertices for each pixel drawn

Something important to keep in mind is that this relates to the OpenGL ES Shading Language, Version 1.00, which is (unfortunately) what’s currently supported by WebGL.

A WebGL implementation must only accept shaders which conform to The OpenGL ES Shading Language, Version 1.00 [GLES20GLSL], and which do not exceed the minimum functionality mandated in Sections 4 and 5 of Appendix A.

Attribute and Varying were part of early versions of, OpenGL-supported, GLSL, but are deprecated as of OpenGL 3.0 / GLSL 1.30.10, and replaced with more generic constructs:

  • in is for input from the previous pipeline stage, i.e. per vertex (or per fragment) values at most, per primitive if using glAttribDivisor and hardware instanciation
  • out is for output to the next stage

Real-time image processing on the web

A while ago I began playing around with grabbing a video stream from a webcam and seeing what I could do with the captured data. Capturing the video stream using the navigator.getUserMedia() method was straightforward, but directly reading and writing the image data of the video stream isn’t possible. That said, the stream data can be put onto a canvas using CanvasRenderingContext2D.drawImage(), giving you to ability to read the pixel data. When it comes to writing visual data, a few options are available.

var videoElem = document.querySelector('video');

// Request video stream
navigator.getUserMedia({video: true, audio: false},

function(_localMediaStream) {
videoStream = _localMediaStream;
videoElem.src = window.URL.createObjectURL(_localMediaStream);
},

function(err) {
console.log(
'navigator.getUserMedia error' + err);
}

);
var videoElem = document.querySelector('video');
var canvas = document.querySelector('canvas');
var ctx = canvas.getContext('2d');

...

// put snapshot from video stream into canvas
ctx.drawImage(videoElem, 0, 0);

You can read and write to the <canvas> element, so hiding the <video> element with the source data and just showing the <canvas> element is an option, but the CanvasRenderingContext2D.drawImage() call is expensive; looking at the copied stream on the <canvas> element there is, very noticeable, visual lag. Another reason to avoid this option is that the frequency at which you render (e.g. 30 FPS), isn’t necessarily the frequency at which you’d want to grab and process image data (e.g. 10 FPS). The disassociation allow you to keep the video playback smooth, for a better user experience, but more effectively utilize CPU cycles for the image processing. At least in my experiences so far, a small delay in the visual feedback from the image processing is acceptable and looks perfectly fine intermixed with the higher-frequency video stream.

Throwing aside reading and writing to just the <canvas> element, alternative options all involve showing the <video> element with the webcam stream and placing visual feedback on top of the video pixels. A few ideas:

  • Write pixel data to another canvas and render it on top of the <video> element
  • Render SVG elements on top of the <video> element
  • Render DOM elements (absolutely positioned) on top of the <video> element

The third option is an ugly solution, but it’s fast to code and thus allows for quick prototyping. The demo and code below shows a quick demo I slapped together using <div> elements as markers for hot spots, in this case bright spots, within the video.

<!DOCTYPE html>
<
html>
<
head>
<
title>Webcam Cap</title>
<
meta charset="UTF-8">
<
meta name="viewport" content="width=device-width, initial-scale=1.0">

<
style type="text/css">
* { margin:0; padding:0; border:none; }
</style>

</
head>

<
body>
<
div>
<
video style="width:640px; height:480px;" width="640" height="480" autoplay></video>
<
canvas style="display:none; width:640px; height:480px;" width="640" height="480"></canvas>
</
div>

<
div class="ia-markers"></div>

<
script type="text/javascript">

navigator.getUserMedia = (navigator.getUserMedia || navigator.webkitGetUserMedia || navigator.mozGetUserMedia || navigator.msGetUserMedia);

if ( typeof navigator.getUserMedia !== 'undefined' ) {

var videoElem = document.querySelector('video');
var canvas = document.querySelector('canvas');
var ctx = canvas.getContext('2d');
var videoStream = null;
var snapshotIntv = null;

var width = 640;
var height = 480;

// Request video stream
navigator.getUserMedia({video: true, audio: false},

function(_localMediaStream) {
videoStream = _localMediaStream;
videoElem.src = window.URL.createObjectURL(_localMediaStream);

// Take a snapshot of the video stream 10ms
snapshotIntv = setInterval(function() {
processSnapshot(videoStream);
}, 100);

},

function(err) {
console.log(
'navigator.getUserMedia error' + err);
}

);


// Take a snapshot from the video stream
function processSnapshot() {

// put snapshot from video stream into canvas
ctx.drawImage(videoElem, 0, 0);

// Clear old snapshot markers
var markerSetParent = (document.getElementsByClassName('ia-markers'))[0];
markerSetParent.innerHTML =
'';

// Array to store hotzone points
var hotzones = [];

// Process pixels
var imageData = ctx.getImageData(0, 0, width, height);
for (var y = 0; y < height; y+=16) {
for (var x = 0; x < width; x+=16) {
var index = (x + y * imageData.width) << 2;

var r = imageData.data[index + 0];
var g = imageData.data[index + 1];
var b = imageData.data[index + 2];

if(r > 200 && g > 200 && b > 200) {
hotzones.push([x,y]);
}
}
}

// Add new hotzone elements to DOM
for(var i=0; i<hotzones.length; i++) {
var x = hotzones[i][0];
var y = hotzones[i][1];

var markerDivElem = document.createElement("div");
markerDivElem.setAttribute(
'style', 'position:absolute; width:16px; height:16px; border-radius:8px; background:#0f0; opacity:0.25; left:' + x + 'px; top:' + y + 'px');
markerDivElem.className =
'ia-hotzone-marker';

markerSetParent.appendChild(markerDivElem);
}
}

}
else {
console.log(
'getUserMedia() is not supported in your browser');
}

</script>

</
body>
</
html>

WebGL on a high-DPI display

Dealing with WebGL on a high-DPI display isn’t too difficult, but it does require an understanding of device pixels vs CSS pixels. Elements on a page automatically upscale on a high-DPI display, as dimensions are typically defined with CSS, and therefore defined in units of CSS pixels. The <canvas> element is no exception. However, upscaling a DOM element doesn’t mean that the content within the element will be upscaled or rendered nicely – this is why non-vector content can appear blurry on higher resolutions. With WebGL content, not only will it appear blurry, but the viewport will likely be clipped as well, due to the viewport being incorrectly calculated using CSS pixel dimensions.

With WebGL everything is assumed to be in units of device pixels and there is no automatic conversion from CSS pixels to device pixels. To specify the device pixel dimensions of the <canvas>, we need to set the width and height attributes of the element:

CSS Pixels vs Device Pixels on canvas

I like to compute and set the attributes automatically using window.devicePixelRatio.
In glfx I do the following (passing in _canvasWidthCSSPx, _canvasHeightCSSPx):

// Get devicePixelRatio
glfx.devicePixelRatio = window.devicePixelRatio || 1;

// Set the width,height attributes of the canvas element (in device pixels)
var _canvasWidthDevicePx = _canvasWidthCSSPx * glfx.devicePixelRatio;
var _canvasHeightDevicePx = _canvasHeightCSSPx * glfx.devicePixelRatio;    
_canvas.setAttribute(
"width", _canvasWidthDevicePx);
_canvas.setAttribute(
"height", _canvasHeightDevicePx);

// Set viewport width,height based on dimensions of canvas element        
glfx.gl.viewportWidth = _canvasWidthDevicePx;
glfx.gl.viewportHeight = _canvasHeightDevicePx;        

Reference: HandlingHighDPI

Where we live

Using a number of technologies I’ve been playing around with recently, I began working on a 3D visualization of the Earth, plotting every city, creating a pointillism-styled representation of the planet. Below is the result along with an overview of how I produced the rendering.

Getting the data

I extracted all cities with a population of at least 100,000 people from the MySQL GeoNames database using the following query:

SELECT `id`,`name`,`latitude`,`longitude`,`population`,`timezone`
FROM geonames.cities
WHERE population >= 100000 AND feature_class = 'P';

… and put the results into a JS array.

Creating a 3D model to represent each city

I created this hexagonal model in Blender, exported it to a Wavefront OBJ file, and ran the OBJ file through the Wavefront OBJ to JSON converter I wrote. Note that the model is facing the z-axis to match WebGL’s (and OpenGL’s) default camera orientation: facing down the negative z-axis.

Convert longitude and latitude to a 3D position

Converting a geodetic longitude, latitude pair to a 3D position involves doing a LLA (Longitude Latitude Altitude) to ECEF (Earth-Centered, Earth-Fixed) transformation. The code below implements this transform, converting the longitude and latitude of every city pulled from the GeoNames database into a 3D coordinate where we can render the hexagonal representation of the city.

function llarToWorld(lat, lon, alt, rad)
{            
    lat = lat * (Math.PI/180.0);
    lon = lon * (Math.PI/180.0);

    
var f = 0; //flattening
    
var ls = Math.atan( Math.pow((1.0 - f),2) * Math.tan(lat) ); // lambda

    
var x = rad * Math.cos(ls) * Math.cos(lon) + alt * Math.cos(lat) * Math.cos(lon)
    
var y = rad * Math.cos(ls) * Math.sin(lon) + alt * Math.cos(lat) * Math.sin(lon)
    
var z = rad * Math.sin(ls) + alt * Math.sin(lat)
    
    
return [x,z,-y];            
}

There are 2 items worth noting:

  • The transformation (and function above) involve a 4th parameter, radius which is the radius of the ellipsoid (or sphere, in this case, as flattening=0) into which the transformation is done. I have it set as a fixed constant, as I’m primary concerned with an approximate visual representation, but the MathWorks page describes the actual computation.
  • The ECEF (Earth-Centered, Earth-Fixed) coordinate system has the z-axis pointing north, not the y-axis, so the z and y values need to be swapped to produce a coordinate corresponding to WebGL’s default camera orientation. In addition, as WebGL has a right-handed coordinate system (so the default camera orientation is one where it’s pointing down the negative z-axis), the z coordinate is negated so the point doesn’t wind up behind the camera.

Orient all cities to face the origin

Getting each of the hexagonal models to face the origin involved a bit of math:

  • Calculating the axis about which the rotation should occur by, first, computing a vector from the origin to the 3D position of the model (lookAt), and taking the cross product between lookAt and the z-axis (as we’re rotating toward the z-axis).
  • Calculating the angle of rotation (the angle between the z-axis and lookAt) by computing the dot product between lookAt and the z-axis, then taking the acos of the dot product.

There’s some additional code to handle cases where points lie on the on the z-axis (where the cross product gives the zero vector) and also to return a matrix representation of the rotation.

function lookAtOrigin(v)
{
// compute vector from origin
var lookAt = vec3.create([v[0], v[1], -v[2]]);
vec3.normalize(lookAt);

// reference axis
var refAxis = vec3.create([0,0,-1]);

// computate axis of rotation
var rotAxis = vec3.create(lookAt);
vec3.cross(rotAxis, refAxis);

// compute angle of rotation
var rotAngRad = Math.acos(vec3.dot(lookAt, refAxis));

// special cases...
if(rotAxis[0] == 0 && rotAxis[1] == 0 && rotAxis[2] == 0) {
if(lookAt[2] > 0) {
rotAxis = vec3.create([1,0,0]);
rotAngRad = Math.PI;
}
else {
rotAxis = vec3.create([1,0,0]);
rotAngRad = 0;
}
}

// compute and return a matrix with the rotation
var ret = mat4.identity();
mat4.rotate(ret, rotAngRad, rotAxis);

return ret;
}

Render the scene

Using glfx, I pulled everything together, also adding a bit of code to rotate the camera and do some pseudo-lighting in the pixel shader by alpha blending colors based on depth. All the code can be found in the webgl-globe repository on bitbucket.

Entering the world of high-DPI displays

With a Retina iPad and my recent purchase of a Yoga 2 laptop with a “Quad HD” display (3200×1800) I’ve been dragged into the world of high-DPI (more precisely PPI) displays. For years, DPI was “standard” at either 96dpi (Windows) or 72dpi (Mac OS), with a logical/software pixel being equivalent to a hardware pixel on the display device. A higher resolution monitor meant the content on your display got a bit smaller but you gained a couple more thousand pixels to work with, but the recent and massive increases in pixel densities seems to be the end of the 1:1 mapping between software and hardware pixel references. Below are a few notes on my experiences dealing with high-DPI displays and content so far.

  • Windows 8.1 support is terrible. Both application support and operating system support for high-DPI displays is abysmal. See the post Living a High-DPI desktop lifestyle can be painful by Scott Hanselman which reflects many of the issues I’ve encountered with my Yoga 2 as well. There’s a large list of issues: for non-DPI aware applications Windows scale text but not icons and layout, applications lie about being DPI-aware and are rendered too small, and some applications simply crash (TourtiseSVN’s diff… no clue why, too many pixels?!). It’s easy to wag a finger at application developers, but legacy support is clearly something the operating system needs to handle. In addition, despite Windows 8.1 touting automatic, per-monitor, DPI detection, it’s all based off of the DPI of a “primary monitor” and content on the other monitors is scaled to match. So dragging a window for an application from the Yoga 2 display across to a HD/96dpi monitor results in the window being scaled down (and visibly blurry). Worse, all applications undergo the same treatment – so if you’re thinking you can just use non-DPI aware applications on an external display until support comes around, guess again.
    ArsTechnica did a piece mentioning this issue in particular, and Windows 8.1’s high-DPI support is general.
  • Web support is only half-way there. The best thing done to support high-DPI displays was defining the CSS2 reference pixel to be independent of hardware pixels. Beyond that you have media queries and higher resolution background images, but there’s still no good way to specify alternate foreground images, though the <picture> element may gain support soon. In general, outside of CSS things gets messy, as is the case with a high-DPI <canvas>.
    One problem with the web that I don’t see a proposed solution for is handling low-resolution image assets for which you can’t get a higher resolution version. This is a problem I face with this blog. There’s a lot of images (old screenshots, low-resolution photos, etc.) for which I can’t get a 2x, 4x, etc., higher resolution version and there’s no way to prevent upscaling the images or specify how the upscaling is done. The typical 2x-bilinear-filtered upscaling, done by most browsers, is not always desirable. In addition, as display vendors pack more pixels in, what happens when the “high-resolution” version needed is 4x or 8x?
  • SVG/Vector-based images aren’t always the answer. There’s a lot of benefits to vector-based formats, but they’re not the holy grail many think they are. For vector-based images, rendering costs grow as you add details with polygons and paths. It’s why video games still rely heavily on texture mapping, even as graphics hardware has progressed to handle rendering millions of polygons per frame – the additional geometry and computation for fine details is enormous.