Can I turn off antialiasing on an HTML <canvas> element?
JavascriptHtmlCanvasAntialiasingJavascript Problem Overview
I'm playing around with the <canvas> element, drawing lines and such.
I've noticed that my diagonal lines are antialiased. I'd prefer the jaggy look for what I'm doing - is there any way of turning this feature off?
Javascript Solutions
Solution 1 - Javascript
For images there's now context.imageSmoothingEnabled = false.
However, there's nothing that explicitly controls line drawing. You may need to draw your own lines (the hard way) using getImageData and putImageData.
Solution 2 - Javascript
Draw your 1-pixel lines on coordinates like ctx.lineTo(10.5, 10.5). Drawing a one-pixel line over the point (10, 10) means, that this 1 pixel at that position reaches from 9.5 to 10.5 which results in two lines that get drawn on the canvas.
A nice trick to not always need to add the 0.5 to the actual coordinate you want to draw over if you've got a lot of one-pixel lines, is to ctx.translate(0.5, 0.5) your whole canvas at the beginning.
Solution 3 - Javascript
It can be done in Mozilla Firefox. Add this to your code:
contextXYZ.mozImageSmoothingEnabled = false;
In Opera it's currently a feature request, but hopefully it will be added soon.
Solution 4 - Javascript
It must antialias vector graphics
Antialiasing is required for correct plotting of vector graphics that involves non-integer coordinates (0.4, 0.4), which all but very few clients do.
When given non-integer coordinates, the canvas has two options:
- Antialias - paint the pixels around the coordinate based on how far the integer coordinate is from non-integer one (ie, the rounding error).
- Round - apply some rounding function to the non-integer coordinate (so 1.4 will become 1, for example).
The later strategy will work for static graphics, although for small graphics (a circle with radius of 2) curves will show clear steps rather than a smooth curve.
The real problem is when the graphics is translated (moved) - the jumps between one pixel and another (1.6 => 2, 1.4 => 1), mean that the origin of the shape may jump with relation to the parent container (constantly shifting 1 pixel up/down and left/right).
Some tips
Tip #1: You can soften (or harden) antialiasing by scaling the canvas (say by x) then apply the reciprocal scale (1/x) to the geometries yourself (not using the canvas).
Compare (no scaling):
with (canvas scale: 0.75; manual scale: 1.33):
and (canvas scale: 1.33; manual scale: 0.75):
Tip #2: If a jaggy look is really what you're after, try to draw each shape a few times (without erasing). With each draw, the antialiasing pixels get darker.
Compare. After drawing once:
After drawing thrice:
Solution 5 - Javascript
I would draw everything using a custom line algorithm such as Bresenham's line algorithm. Check out this javascript implementation: http://members.chello.at/easyfilter/canvas.html
I think this will definitely solve your problems.
Solution 6 - Javascript
I want to add that I had trouble when downsizing an image and drawing on canvas, it was still using smoothing, even though it wasn't using when upscaling.
I solved using this:
function setpixelated(context){
context['imageSmoothingEnabled'] = false; /* standard */
context['mozImageSmoothingEnabled'] = false; /* Firefox */
context['oImageSmoothingEnabled'] = false; /* Opera */
context['webkitImageSmoothingEnabled'] = false; /* Safari */
context['msImageSmoothingEnabled'] = false; /* IE */
}
You can use this function like this:
var canvas = document.getElementById('mycanvas')
setpixelated(canvas.getContext('2d'))
Maybe this is useful for someone.
Solution 7 - Javascript
ctx.translate(0.5, 0.5);
ctx.lineWidth = .5;
With this combo I can draw nice 1px thin lines.
Solution 8 - Javascript
Try something like canvas { image-rendering: pixelated; }.
This might not work if you're trying to only make one line not antialiased.
const canvas = document.querySelector("canvas");
const ctx = canvas.getContext("2d");
ctx.fillRect(4, 4, 2, 2);
canvas {
image-rendering: pixelated;
width: 100px;
height: 100px; /* Scale 10x */
}
<html>
<head></head>
<body>
<canvas width="10" height="10">Canvas unsupported</canvas>
</body>
</html>
I haven't tested this on many browsers though.
Solution 9 - Javascript
Adding this:
image-rendering: pixelated; image-rendering: crisp-edges;
to the style attribute of the canvas element helped to draw crisp pixels on the canvas. Discovered via this great article:
https://developer.mozilla.org/en-US/docs/Games/Techniques/Crisp_pixel_art_look
Solution 10 - Javascript
Notice a very limited trick. If you want to create a 2 colors image, you may draw any shape you want with color #010101 on a background with color #000000. Once this is done, you may test each pixel in the imageData.data[] and set to 0xFF whatever value is not 0x00 :
imageData = context2d.getImageData (0, 0, g.width, g.height);
for (i = 0; i != imageData.data.length; i ++) {
if (imageData.data[i] != 0x00)
imageData.data[i] = 0xFF;
}
context2d.putImageData (imageData, 0, 0);
The result will be a non-antialiased black & white picture. This will not be perfect, since some antialiasing will take place, but this antialiasing will be very limited, the color of the shape being very much like the color of the background.
Solution 11 - Javascript
I discovered a better way to disable antialiasing on path / shape rendering using the context's filter property:
The magic / TL;DR:
ctx = canvas.getContext('2d');
// make canvas context render without antialiasing
ctx.filter = "url(data:image/svg+xml;base64,PHN2ZyB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciPjxmaWx0ZXIgaWQ9ImZpbHRlciIgeD0iMCIgeT0iMCIgd2lkdGg9IjEwMCUiIGhlaWdodD0iMTAwJSIgY29sb3ItaW50ZXJwb2xhdGlvbi1maWx0ZXJzPSJzUkdCIj48ZmVDb21wb25lbnRUcmFuc2Zlcj48ZmVGdW5jUiB0eXBlPSJpZGVudGl0eSIvPjxmZUZ1bmNHIHR5cGU9ImlkZW50aXR5Ii8+PGZlRnVuY0IgdHlwZT0iaWRlbnRpdHkiLz48ZmVGdW5jQSB0eXBlPSJkaXNjcmV0ZSIgdGFibGVWYWx1ZXM9IjAgMSIvPjwvZmVDb21wb25lbnRUcmFuc2Zlcj48L2ZpbHRlcj48L3N2Zz4=#filter)";
Demystified:
The data url is a reference to an SVG containing a single filter:
<svg xmlns="http://www.w3.org/2000/svg">
<filter id="filter" x="0" y="0" width="100%" height="100%" color-interpolation-filters="sRGB">
<feComponentTransfer>
<feFuncR type="identity"/>
<feFuncG type="identity"/>
<feFuncB type="identity"/>
<feFuncA type="discrete" tableValues="0 1"/>
</feComponentTransfer>
</filter>
</svg>
Then at the very end of the url is an id reference to that #filter:
"url(data:image/svg+...Zz4=#filter)";
The SVG filter uses a discrete transform on the alpha channel, selecting only completely transparent or completely opaque on a 50% boundary when rendering. This can be tweaked to add some anti-aliasing back in if needed, e.g.:
...
<feFuncA type="discrete" tableValues="0 0 0.25 0.75 1"/>
...
Cons / Notes / Gotchas
Note, I didn't test this method with images, but I can presume it would affect semi-transparent parts of images. I can also guess that it probably would not prevent antialiasing on images at differing color boundaries. It isn't a 'nearest color' solution but rather a binary transparency solution. It seems to work best with path / shape rendering since alpha is the only channel antialiased with paths.
Also, using a minimum lineWidth of 1 is safe. Thinner lines become sparse or may often disappear completely.
Edit:
I've discovered that, in Firefox, setting filter to a dataurl does not work immediately / synchronously: the dataurl has to 'load' first.
e.g. The following will not work in Firefox:
ctx.filter = "url(data:image/svg+xml;base64,...#filter)";
ctx.beginPath();
ctx.moveTo(10,10);
ctx.lineTo(20,20);
ctx.strokeStyle = 'black';
ctx.lineWidth = 2;
ctx.stroke();
ctx.filter = "none";
But waiting till the next JS frame works fine:
ctx.filter = "url(data:image/svg+xml;base64,...#filter)";
setTimeout(() => {
ctx.beginPath();
ctx.moveTo(10,10);
ctx.lineTo(20,20);
ctx.strokeStyle = 'black';
ctx.lineWidth = 2;
ctx.stroke();
ctx.filter = "none";
}, 0);
Solution 12 - Javascript
Here is a basic implementation of Bresenham's algorithm in JavaScript. It's based on the integer-arithmetic version described in this wikipedia article: https://en.wikipedia.org/wiki/Bresenham%27s_line_algorithm
function range(f=0, l) {
var list = [];
const lower = Math.min(f, l);
const higher = Math.max(f, l);
for (var i = lower; i <= higher; i++) {
list.push(i);
}
return list;
}
//Don't ask me.
//https://en.wikipedia.org/wiki/Bresenham%27s_line_algorithm
function bresenhamLinePoints(start, end) {
let points = [];
if(start.x === end.x) {
return range(f=start.y, l=end.y)
.map(yIdx => {
return {x: start.x, y: yIdx};
});
} else if (start.y === end.y) {
return range(f=start.x, l=end.x)
.map(xIdx => {
return {x: xIdx, y: start.y};
});
}
let dx = Math.abs(end.x - start.x);
let sx = start.x < end.x ? 1 : -1;
let dy = -1*Math.abs(end.y - start.y);
let sy = start.y < end.y ? 1 : - 1;
let err = dx + dy;
let currX = start.x;
let currY = start.y;
while(true) {
points.push({x: currX, y: currY});
if(currX === end.x && currY === end.y) break;
let e2 = 2*err;
if (e2 >= dy) {
err += dy;
currX += sx;
}
if(e2 <= dx) {
err += dx;
currY += sy;
}
}
return points;
}
Solution 13 - Javascript
While we still don't have proper shapeSmoothingEnabled or shapeSmoothingQuality options on the 2D context (I'll advocate for this and hope it makes its way in the near future), we now have ways to approximate a "no-antialiasing" behavior, thanks to SVGFilters, which can be applied to the context through its .filter property.
So, to be clear, it won't deactivate antialiasing per se, but provides a cheap way both in term of implementation and of performances (?, it should be hardware accelerated, which should be better than a home-made Bresenham on the CPU) in order to remove all semi-transparent pixels while drawing, but it may also create some blobs of pixels, and may not preserve the original input color.
For this we can use a <feComponentTransfer> node to grab only fully opaque pixels.
const canvas = document.getElementById("canvas");
const ctx = canvas.getContext("2d");
ctx.fillStyle = "#ABEDBE";
ctx.fillRect(0,0,canvas.width,canvas.height);
ctx.fillStyle = "black";
ctx.font = "14px sans-serif";
ctx.textAlign = "center";
// first without filter
ctx.fillText("no filter", 60, 20);
drawArc();
drawTriangle();
// then with filter
ctx.setTransform(1, 0, 0, 1, 120, 0);
ctx.filter = "url(#remove-alpha)";
// and do the same ops
ctx.fillText("no alpha", 60, 20);
drawArc();
drawTriangle();
// to remove the filter
ctx.filter = "none";
function drawArc() {
ctx.beginPath();
ctx.arc(60, 80, 50, 0, Math.PI * 2);
ctx.stroke();
}
function drawTriangle() {
ctx.beginPath();
ctx.moveTo(60, 150);
ctx.lineTo(110, 230);
ctx.lineTo(10, 230);
ctx.closePath();
ctx.stroke();
}
// unrelated
// simply to show a zoomed-in version
const zoomed = document.getElementById("zoomed");
const zCtx = zoomed.getContext("2d");
zCtx.imageSmoothingEnabled = false;
canvas.onmousemove = function drawToZoommed(e) {
const
x = e.pageX - this.offsetLeft,
y = e.pageY - this.offsetTop,
w = this.width,
h = this.height;
zCtx.clearRect(0,0,w,h);
zCtx.drawImage(this, x-w/6,y-h/6,w, h, 0,0,w*3, h*3);
}
<svg width="0" height="0" style="position:absolute;z-index:-1;">
<defs>
<filter id="remove-alpha" x="0" y="0" width="100%" height="100%">
<feComponentTransfer>
<feFuncA type="discrete" tableValues="0 1"></feFuncA>
</feComponentTransfer>
</filter>
</defs>
</svg>
<canvas id="canvas" width="250" height="250" ></canvas>
<canvas id="zoomed" width="250" height="250" ></canvas>
For the ones that don't like to append an <svg> element in their DOM, and who live in the near future (or with experimental flags on), the CanvasFilter interface we're working on should allow to do this without a DOM (so from Worker too):
if (!("CanvasFilter" in globalThis)) {
throw new Error("Not Supported", "Please enable experimental web platform features, or wait a bit");
}
const canvas = document.getElementById("canvas");
const ctx = canvas.getContext("2d");
ctx.fillStyle = "#ABEDBE";
ctx.fillRect(0,0,canvas.width,canvas.height);
ctx.fillStyle = "black";
ctx.font = "14px sans-serif";
ctx.textAlign = "center";
// first without filter
ctx.fillText("no filter", 60, 20);
drawArc();
drawTriangle();
// then with filter
ctx.setTransform(1, 0, 0, 1, 120, 0);
ctx.filter = new CanvasFilter([
{
filter: "componentTransfer",
funcA: {
type: "discrete",
tableValues: [ 0, 1 ]
}
}
]);
// and do the same ops
ctx.fillText("no alpha", 60, 20);
drawArc();
drawTriangle();
// to remove the filter
ctx.filter = "none";
function drawArc() {
ctx.beginPath();
ctx.arc(60, 80, 50, 0, Math.PI * 2);
ctx.stroke();
}
function drawTriangle() {
ctx.beginPath();
ctx.moveTo(60, 150);
ctx.lineTo(110, 230);
ctx.lineTo(10, 230);
ctx.closePath();
ctx.stroke();
}
// unrelated
// simply to show a zoomed-in version
const zoomed = document.getElementById("zoomed");
const zCtx = zoomed.getContext("2d");
zCtx.imageSmoothingEnabled = false;
canvas.onmousemove = function drawToZoommed(e) {
const
x = e.pageX - this.offsetLeft,
y = e.pageY - this.offsetTop,
w = this.width,
h = this.height;
zCtx.clearRect(0,0,w,h);
zCtx.drawImage(this, x-w/6,y-h/6,w, h, 0,0,w*3, h*3);
};
<canvas id="canvas" width="250" height="250" ></canvas>
<canvas id="zoomed" width="250" height="250" ></canvas>
Or you can also save the SVG as an external file and set the filter property to path/to/svg_file.svg#remove-alpha.
Solution 14 - Javascript
For those who still looking for answers. here is my solution.
Assumming image is 1 channel gray. I just thresholded after ctx.stroke().
ctx.beginPath();
ctx.moveTo(some_x, some_y);
ctx.lineTo(some_x, some_y);
...
ctx.closePath();
ctx.fill();
ctx.stroke();
let image = ctx.getImageData(0, 0, ctx.canvas.width, ctx.canvas.height)
for(let x=0; x < ctx.canvas.width; x++) {
for(let y=0; y < ctx.canvas.height; y++) {
if(image.data[x*image.height + y] < 128) {
image.data[x*image.height + y] = 0;
} else {
image.data[x*image.height + y] = 255;
}
}
}
if your image channel is 3 or 4. you need to modify the array index like
x*image.height*number_channel + y*number_channel + channel
Solution 15 - Javascript
Just two notes on StashOfCode's answer:
- It only works for a grayscale, opaque canvas (fillRect with white then draw with black, or viceversa)
- It may fail when lines are thin (~1px line width)
It's better to do this instead:
Stroke and fill with #FFFFFF, then do this:
imageData.data[i] = (imageData.data[i] >> 7) * 0xFF
That solves it for lines with 1px width.
Other than that, StashOfCode's solution is perfect because it doesn't require to write your own rasterization functions (think not only lines but beziers, circular arcs, filled polygons with holes, etc...)