javascript - ratio - resizing and cropping images with canvas




Resizing an image in an HTML5 canvas (12)

I'm trying to create a thumbnail image on the client side using javascript and a canvas element, but when I shrink the image down, it looks terrible. It looks as if it was downsized in photoshop with the resampling set to 'Nearest Neighbor' instead of Bicubic. I know its possible to get this to look right, because this site can do it just fine using a canvas as well. I've tried using the same code they do as shown in the "[Source]" link, but it still looks terrible. Is there something I'm missing, some setting that needs to be set or something?

EDIT:

I'm trying to resize a jpg. I have tried resizing the same jpg on the linked site and in photoshop, and it looks fine when downsized.

Here is the relevant code:

reader.onloadend = function(e)
{
    var img = new Image();
    var ctx = canvas.getContext("2d");
    var canvasCopy = document.createElement("canvas");
    var copyContext = canvasCopy.getContext("2d");

    img.onload = function()
    {
        var ratio = 1;

        if(img.width > maxWidth)
            ratio = maxWidth / img.width;
        else if(img.height > maxHeight)
            ratio = maxHeight / img.height;

        canvasCopy.width = img.width;
        canvasCopy.height = img.height;
        copyContext.drawImage(img, 0, 0);

        canvas.width = img.width * ratio;
        canvas.height = img.height * ratio;
        ctx.drawImage(canvasCopy, 0, 0, canvasCopy.width, canvasCopy.height, 0, 0, canvas.width, canvas.height);
    };

    img.src = reader.result;
}

EDIT2:

Seems I was mistaken, the linked website wasn't doing any better of a job of downsizing the image. I tried the other methods suggested and none of them look any better. This is what the different methods resulted in:

Photoshop:

Canvas:

Image with image-rendering: optimizeQuality set and scaled with width/height:

Image with image-rendering: optimizeQuality set and scaled with -moz-transform:

Canvas resize on pixastic:

I guess this means firefox isn't using bicubic sampling like its supposed to. I'll just have to wait until they actually add it.

EDIT3:

Original Image


Fast and simple Javascript image resizer:

https://github.com/calvintwr/Hermite-resize

Use:

h.resize({
    source: document.getElementById('image'), // any canvas or image elements, jQuery or native
    width: 400,
    height: 600,
    output: 'image', // [optional] `image` or `canvas`. If not entered output is same as input element.
    quality: 0.7, // [optional] applicable for `image` output only
}, function(output) {
    //your callback
});

History

This is really after many rounds of research, reading and trying.

The resizer algorithm uses @ViliusL's Hermite script (Hermite resizer is really the fastest and gives reasonably good output). Extended with features you need.

Forks 1 worker to do the resizing so that it doesn't freeze your browser when resizing, unlike all other JS resizers out there.


Fast image resize/resample algorithm using Hermite filter with JavaScript. Support transparency, gives good quality. Preview:


Update: version 2.0 added on GitHub (faster, web workers + transferable objects). Finally i got it working!

Git: https://github.com/viliusle/Hermite-resize
Demo: http://viliusle.github.io/miniPaint/

/**
 * Hermite resize - fast image resize/resample using Hermite filter. 1 cpu version!
 * 
 * @param {HtmlElement} canvas
 * @param {int} width
 * @param {int} height
 * @param {boolean} resize_canvas if true, canvas will be resized. Optional.
 */
function resample_single(canvas, width, height, resize_canvas) {
    var width_source = canvas.width;
    var height_source = canvas.height;
    width = Math.round(width);
    height = Math.round(height);

    var ratio_w = width_source / width;
    var ratio_h = height_source / height;
    var ratio_w_half = Math.ceil(ratio_w / 2);
    var ratio_h_half = Math.ceil(ratio_h / 2);

    var ctx = canvas.getContext("2d");
    var img = ctx.getImageData(0, 0, width_source, height_source);
    var img2 = ctx.createImageData(width, height);
    var data = img.data;
    var data2 = img2.data;

    for (var j = 0; j < height; j++) {
        for (var i = 0; i < width; i++) {
            var x2 = (i + j * width) * 4;
            var weight = 0;
            var weights = 0;
            var weights_alpha = 0;
            var gx_r = 0;
            var gx_g = 0;
            var gx_b = 0;
            var gx_a = 0;
            var center_y = (j + 0.5) * ratio_h;
            var yy_start = Math.floor(j * ratio_h);
            var yy_stop = Math.ceil((j + 1) * ratio_h);
            for (var yy = yy_start; yy < yy_stop; yy++) {
                var dy = Math.abs(center_y - (yy + 0.5)) / ratio_h_half;
                var center_x = (i + 0.5) * ratio_w;
                var w0 = dy * dy; //pre-calc part of w
                var xx_start = Math.floor(i * ratio_w);
                var xx_stop = Math.ceil((i + 1) * ratio_w);
                for (var xx = xx_start; xx < xx_stop; xx++) {
                    var dx = Math.abs(center_x - (xx + 0.5)) / ratio_w_half;
                    var w = Math.sqrt(w0 + dx * dx);
                    if (w >= 1) {
                        //pixel too far
                        continue;
                    }
                    //hermite filter
                    weight = 2 * w * w * w - 3 * w * w + 1;
                    var pos_x = 4 * (xx + yy * width_source);
                    //alpha
                    gx_a += weight * data[pos_x + 3];
                    weights_alpha += weight;
                    //colors
                    if (data[pos_x + 3] < 255)
                        weight = weight * data[pos_x + 3] / 250;
                    gx_r += weight * data[pos_x];
                    gx_g += weight * data[pos_x + 1];
                    gx_b += weight * data[pos_x + 2];
                    weights += weight;
                }
            }
            data2[x2] = gx_r / weights;
            data2[x2 + 1] = gx_g / weights;
            data2[x2 + 2] = gx_b / weights;
            data2[x2 + 3] = gx_a / weights_alpha;
        }
    }
    //clear and resize canvas
    if (resize_canvas === true) {
        canvas.width = width;
        canvas.height = height;
    } else {
        ctx.clearRect(0, 0, width_source, height_source);
    }

    //draw
    ctx.putImageData(img2, 0, 0);
}

I converted @syockit's answer as well as the step-down approach into a reusable Angular service for anyone who's interested: https://gist.github.com/fisch0920/37bac5e741eaec60e983

I included both solutions because they both have their own pros / cons. The lanczos convolution approach is higher quality at the cost of being slower, whereas the step-wise downscaling approach produces reasonably antialiased results and is significantly faster.

Example usage:

angular.module('demo').controller('ExampleCtrl', function (imageService) {
  // EXAMPLE USAGE
  // NOTE: it's bad practice to access the DOM inside a controller, 
  // but this is just to show the example usage.

  // resize by lanczos-sinc filter
  imageService.resize($('#myimg')[0], 256, 256)
    .then(function (resizedImage) {
      // do something with resized image
    })

  // resize by stepping down image size in increments of 2x
  imageService.resizeStep($('#myimg')[0], 256, 256)
    .then(function (resizedImage) {
      // do something with resized image
    })
})

I have a feeling the module I wrote will produce similar results to photoshop, as it preserves color data by averaging them, not applying an algorithm. It's kind of slow, but to me it is the best, because it preserves all the color data.

https://github.com/danschumann/limby-resize/blob/master/lib/canvas_resize.js

It doesn't take the nearest neighbor and drop other pixels, or sample a group and take a random average. It takes the exact proportion each source pixel should output into the destination pixel. The average pixel color in the source will be the average pixel color in the destination, which these other formulas, I think they will not be.

an example of how to use is at the bottom of https://github.com/danschumann/limby-resize

UPDATE OCT 2018: These days my example is more academic than anything else. Webgl is pretty much 100%, so you'd be better off resizing with that to produce similar results, but faster. PICA.js does this, I believe. –


I know this is an old thread but it might be useful for some people such as myself that months after are hitting this issue for the first time.

Here is some code that resizes the image every time you reload the image. I am aware this is not optimal at all, but I provide it as a proof of concept.

Also, sorry for using jQuery for simple selectors but I just feel too comfortable with the syntax.

$(document).on('ready', createImage);
$(window).on('resize', createImage);

var createImage = function(){
  var canvas = document.getElementById('myCanvas');
  canvas.width = window.innerWidth || $(window).width();
  canvas.height = window.innerHeight || $(window).height();
  var ctx = canvas.getContext('2d');
  img = new Image();
  img.addEventListener('load', function () {
    ctx.drawImage(this, 0, 0, w, h);
  });
  img.src = 'http://www.ruinvalor.com/Telanor/images/original.jpg';
};
html, body{
  height: 100%;
  width: 100%;
  margin: 0;
  padding: 0;
  background: #000;
}
canvas{
  position: absolute;
  left: 0;
  top: 0;
  z-index: 0;
}
<script src="https://ajax.googleapis.com/ajax/libs/jquery/2.1.1/jquery.min.js"></script>
<html>
  <head>
    <meta charset="utf-8" />
    <title>Canvas Resize</title>
  </head>
  <body>
    <canvas id="myCanvas"></canvas>
  </body>
</html>

My createImage function is called once when the document is loaded and after that it is called every time the window receives a resize event.

I tested it in Chrome 6 and Firefox 3.6, both on the Mac. This "technique" eats processor as it if was ice cream in the summer, but it does the trick.


I'd highly suggest you check out this link and make sure it is set to true.

Controlling image scaling behavior

Introduced in Gecko 1.9.2 (Firefox 3.6 / Thunderbird 3.1 / Fennec 1.0)

Gecko 1.9.2 introduced the mozImageSmoothingEnabled property to the canvas element; if this Boolean value is false, images won't be smoothed when scaled. This property is true by default. view plainprint?

  1. cx.mozImageSmoothingEnabled = false;

If you're simply trying to resize an image, I'd recommend setting width and height of the image with CSS. Here's a quick example:

.small-image {
    width: 100px;
    height: 100px;
}

Note that the height and width can also be set using JavaScript. Here's quick code sample:

var img = document.getElement("my-image");
img.style.width = 100 + "px";  // Make sure you add the "px" to the end,
img.style.height = 100 + "px"; // otherwise you'll confuse IE

Also, to ensure that the resized image looks good, add the following css rules to image selector:

As far as I can tell, all browsers except IE using an bicubic algorithm to resize images by default, so your resized images should look good in Firefox and Chrome.

If setting the css width and height doesn't work, you may want to play with a css transform:

If for whatever reason you need to use a canvas, please note that there are two ways an image can be resize: by resizing the canvas with css or by drawing the image at a smaller size.

See this question for more details.

Hope this helps!


Looking for another great simple solution?

var img=document.createElement('img');
img.src=canvas.toDataURL();
$(img).css("background", backgroundColor);
$(img).width(settings.width);
$(img).height(settings.height);

This solution will use the resize algorith of browser! :)


So what do you do if all the browsers (actually, Chrome 5 gave me quite good one) won't give you good enough resampling quality? You implement them yourself then! Oh come on, we're entering the new age of Web 3.0, HTML5 compliant browsers, super optimized JIT javascript compilers, multi-core(†) machines, with tons of memory, what are you afraid of? Hey, there's the word java in javascript, so that should guarantee the performance, right? Behold, the thumbnail generating code:

// returns a function that calculates lanczos weight
function lanczosCreate(lobes) {
    return function(x) {
        if (x > lobes)
            return 0;
        x *= Math.PI;
        if (Math.abs(x) < 1e-16)
            return 1;
        var xx = x / lobes;
        return Math.sin(x) * Math.sin(xx) / x / xx;
    };
}

// elem: canvas element, img: image element, sx: scaled width, lobes: kernel radius
function thumbnailer(elem, img, sx, lobes) {
    this.canvas = elem;
    elem.width = img.width;
    elem.height = img.height;
    elem.style.display = "none";
    this.ctx = elem.getContext("2d");
    this.ctx.drawImage(img, 0, 0);
    this.img = img;
    this.src = this.ctx.getImageData(0, 0, img.width, img.height);
    this.dest = {
        width : sx,
        height : Math.round(img.height * sx / img.width),
    };
    this.dest.data = new Array(this.dest.width * this.dest.height * 3);
    this.lanczos = lanczosCreate(lobes);
    this.ratio = img.width / sx;
    this.rcp_ratio = 2 / this.ratio;
    this.range2 = Math.ceil(this.ratio * lobes / 2);
    this.cacheLanc = {};
    this.center = {};
    this.icenter = {};
    setTimeout(this.process1, 0, this, 0);
}

thumbnailer.prototype.process1 = function(self, u) {
    self.center.x = (u + 0.5) * self.ratio;
    self.icenter.x = Math.floor(self.center.x);
    for (var v = 0; v < self.dest.height; v++) {
        self.center.y = (v + 0.5) * self.ratio;
        self.icenter.y = Math.floor(self.center.y);
        var a, r, g, b;
        a = r = g = b = 0;
        for (var i = self.icenter.x - self.range2; i <= self.icenter.x + self.range2; i++) {
            if (i < 0 || i >= self.src.width)
                continue;
            var f_x = Math.floor(1000 * Math.abs(i - self.center.x));
            if (!self.cacheLanc[f_x])
                self.cacheLanc[f_x] = {};
            for (var j = self.icenter.y - self.range2; j <= self.icenter.y + self.range2; j++) {
                if (j < 0 || j >= self.src.height)
                    continue;
                var f_y = Math.floor(1000 * Math.abs(j - self.center.y));
                if (self.cacheLanc[f_x][f_y] == undefined)
                    self.cacheLanc[f_x][f_y] = self.lanczos(Math.sqrt(Math.pow(f_x * self.rcp_ratio, 2)
                            + Math.pow(f_y * self.rcp_ratio, 2)) / 1000);
                weight = self.cacheLanc[f_x][f_y];
                if (weight > 0) {
                    var idx = (j * self.src.width + i) * 4;
                    a += weight;
                    r += weight * self.src.data[idx];
                    g += weight * self.src.data[idx + 1];
                    b += weight * self.src.data[idx + 2];
                }
            }
        }
        var idx = (v * self.dest.width + u) * 3;
        self.dest.data[idx] = r / a;
        self.dest.data[idx + 1] = g / a;
        self.dest.data[idx + 2] = b / a;
    }

    if (++u < self.dest.width)
        setTimeout(self.process1, 0, self, u);
    else
        setTimeout(self.process2, 0, self);
};
thumbnailer.prototype.process2 = function(self) {
    self.canvas.width = self.dest.width;
    self.canvas.height = self.dest.height;
    self.ctx.drawImage(self.img, 0, 0, self.dest.width, self.dest.height);
    self.src = self.ctx.getImageData(0, 0, self.dest.width, self.dest.height);
    var idx, idx2;
    for (var i = 0; i < self.dest.width; i++) {
        for (var j = 0; j < self.dest.height; j++) {
            idx = (j * self.dest.width + i) * 3;
            idx2 = (j * self.dest.width + i) * 4;
            self.src.data[idx2] = self.dest.data[idx];
            self.src.data[idx2 + 1] = self.dest.data[idx + 1];
            self.src.data[idx2 + 2] = self.dest.data[idx + 2];
        }
    }
    self.ctx.putImageData(self.src, 0, 0);
    self.canvas.style.display = "block";
};

...with which you can produce results like these!

so anyway, here is a 'fixed' version of your example:

img.onload = function() {
    var canvas = document.createElement("canvas");
    new thumbnailer(canvas, img, 188, 3); //this produces lanczos3
    // but feel free to raise it up to 8. Your client will appreciate
    // that the program makes full use of his machine.
    document.body.appendChild(canvas);
};

Now it's time to pit your best browsers out there and see which one will least likely increase your client's blood pressure!

Umm, where's my sarcasm tag?

(since many parts of the code is based on Anrieff Gallery Generator is it also covered under GPL2? I dunno)

actually due to limitation of javascript, multi-core is not supported.


Thanks @syockit for an awesome answer. however, I had to reformat a little as follows to make it work. Perhaps due to DOM scanning issues:

$(document).ready(function () {

$('img').on("load", clickA);
function clickA() {
    var img = this;
    var canvas = document.createElement("canvas");
    new thumbnailer(canvas, img, 50, 3);
    document.body.appendChild(canvas);
}

function thumbnailer(elem, img, sx, lobes) {
    this.canvas = elem;
    elem.width = img.width;
    elem.height = img.height;
    elem.style.display = "none";
    this.ctx = elem.getContext("2d");
    this.ctx.drawImage(img, 0, 0);
    this.img = img;
    this.src = this.ctx.getImageData(0, 0, img.width, img.height);
    this.dest = {
        width: sx,
        height: Math.round(img.height * sx / img.width)
    };
    this.dest.data = new Array(this.dest.width * this.dest.height * 3);
    this.lanczos = lanczosCreate(lobes);
    this.ratio = img.width / sx;
    this.rcp_ratio = 2 / this.ratio;
    this.range2 = Math.ceil(this.ratio * lobes / 2);
    this.cacheLanc = {};
    this.center = {};
    this.icenter = {};
    setTimeout(process1, 0, this, 0);
}

//returns a function that calculates lanczos weight
function lanczosCreate(lobes) {
    return function (x) {
        if (x > lobes)
            return 0;
        x *= Math.PI;
        if (Math.abs(x) < 1e-16)
            return 1
        var xx = x / lobes;
        return Math.sin(x) * Math.sin(xx) / x / xx;
    }
}

process1 = function (self, u) {
    self.center.x = (u + 0.5) * self.ratio;
    self.icenter.x = Math.floor(self.center.x);
    for (var v = 0; v < self.dest.height; v++) {
        self.center.y = (v + 0.5) * self.ratio;
        self.icenter.y = Math.floor(self.center.y);
        var a, r, g, b;
        a = r = g = b = 0;
        for (var i = self.icenter.x - self.range2; i <= self.icenter.x + self.range2; i++) {
            if (i < 0 || i >= self.src.width)
                continue;
            var f_x = Math.floor(1000 * Math.abs(i - self.center.x));
            if (!self.cacheLanc[f_x])
                self.cacheLanc[f_x] = {};
            for (var j = self.icenter.y - self.range2; j <= self.icenter.y + self.range2; j++) {
                if (j < 0 || j >= self.src.height)
                    continue;
                var f_y = Math.floor(1000 * Math.abs(j - self.center.y));
                if (self.cacheLanc[f_x][f_y] == undefined)
                    self.cacheLanc[f_x][f_y] = self.lanczos(Math.sqrt(Math.pow(f_x * self.rcp_ratio, 2) + Math.pow(f_y * self.rcp_ratio, 2)) / 1000);
                weight = self.cacheLanc[f_x][f_y];
                if (weight > 0) {
                    var idx = (j * self.src.width + i) * 4;
                    a += weight;
                    r += weight * self.src.data[idx];
                    g += weight * self.src.data[idx + 1];
                    b += weight * self.src.data[idx + 2];
                }
            }
        }
        var idx = (v * self.dest.width + u) * 3;
        self.dest.data[idx] = r / a;
        self.dest.data[idx + 1] = g / a;
        self.dest.data[idx + 2] = b / a;
    }

    if (++u < self.dest.width)
        setTimeout(process1, 0, self, u);
    else
        setTimeout(process2, 0, self);
};

process2 = function (self) {
    self.canvas.width = self.dest.width;
    self.canvas.height = self.dest.height;
    self.ctx.drawImage(self.img, 0, 0);
    self.src = self.ctx.getImageData(0, 0, self.dest.width, self.dest.height);
    var idx, idx2;
    for (var i = 0; i < self.dest.width; i++) {
        for (var j = 0; j < self.dest.height; j++) {
            idx = (j * self.dest.width + i) * 3;
            idx2 = (j * self.dest.width + i) * 4;
            self.src.data[idx2] = self.dest.data[idx];
            self.src.data[idx2 + 1] = self.dest.data[idx + 1];
            self.src.data[idx2 + 2] = self.dest.data[idx + 2];
        }
    }
    self.ctx.putImageData(self.src, 0, 0);
    self.canvas.style.display = "block";
}
});

This is a javascript function adapted from @Telanor's code. When passing a image base64 as first argument to the function, it returns the base64 of the resized image. maxWidth and maxHeight are optional.

function thumbnail(base64, maxWidth, maxHeight) {

  // Max size for thumbnail
  if(typeof(maxWidth) === 'undefined') var maxWidth = 500;
  if(typeof(maxHeight) === 'undefined') var maxHeight = 500;

  // Create and initialize two canvas
  var canvas = document.createElement("canvas");
  var ctx = canvas.getContext("2d");
  var canvasCopy = document.createElement("canvas");
  var copyContext = canvasCopy.getContext("2d");

  // Create original image
  var img = new Image();
  img.src = base64;

  // Determine new ratio based on max size
  var ratio = 1;
  if(img.width > maxWidth)
    ratio = maxWidth / img.width;
  else if(img.height > maxHeight)
    ratio = maxHeight / img.height;

  // Draw original image in second canvas
  canvasCopy.width = img.width;
  canvasCopy.height = img.height;
  copyContext.drawImage(img, 0, 0);

  // Copy and resize second canvas to first canvas
  canvas.width = img.width * ratio;
  canvas.height = img.height * ratio;
  ctx.drawImage(canvasCopy, 0, 0, canvasCopy.width, canvasCopy.height, 0, 0, canvas.width, canvas.height);

  return canvas.toDataURL();

}

Try pica - that's a highly optimized resizer with selectable algorythms. See demo.

For example, original image from first post is resized in 120ms with Lanczos filter and 3px window or 60ms with Box filter and 0.5px window. For huge 17mb image 5000x3000px resize takes ~1s on desktop and 3s on mobile.

All resize principles were described very well in this thread, and pica does not add rocket science. But it's optimized very well for modern JIT-s, and is ready to use out of box (via npm or bower). Also, it use webworkers when available to avoid interface freezes.

I also plan to add unsharp mask support soon, because it's very useful after downscale.







image-resizing