2
votes

I am using the HTML5 canvas API to draw a tile map for a pixel art game. The rendered tile map is comprised of many smaller images that are cut out of a single source image called a tile sheet. I am using drawImage(src_img, sx, sy, sw, sh, dx, dy, dw, dh) to cut the individual tiles out of the source image and draw them onto the destination canvas. I am using setTransform(sx, 0, 0, sy, tx, ty) to apply scale and translation to the final rendered image.

The color "bleeding" issue I need to fix is caused by the sampler, which uses interpolation to blend colors during scale operations in order to make things not look pixelated. This is great for scaling digital photographs, but not for pixel art. While this doesn't do much visual damage to the centers of the tiles, the sampler is blending colors along the edges of adjacent tiles in the source image which creates unexpected colors in the rendered tile map. Instead of only using colors that fall within the source rectangle passed to drawImage, the sampler blends in colors from just outside of its boundaries causing what appear to be gaps between the tiles.

Below is my tile sheet's source image. Its actual size is 24x24 pixels, but I scaled it up to 96x96 pixels in GIMP so you could see it. I used the "Interpolation: None" setting on GIMP's scaling tool. As you can see there are no gaps or blurred borders around the individual tiles because the sampler did not interpolate the colors. The canvas API's sampler apparently does interpolate colors even when imageSmoothingEnabled is set to false.

enter image description here

Below is a section of the rendered tile map with imageSmoothingEnabled set to true. The left arrow points to some red bleeding into the bottom of the gray tile. This is because the red tile is directly below the gray tile in the tile sheet. The sampler is blending the red into the bottom edge of the gray tile.

The arrow on the right points to the right edge of the green tile. As you can see, no color is bleeding into it. This is because there is nothing to the right of the green tile in the source image and therefore nothing for the sampler to blend.

Tile Map With Image Smoothing Enabled

Below is the rendered tile map with imageSmoothingEnabled set to false. Depending on the scale and translation, texture bleeding still occurs. The left arrow is pointing to red bleeding in from the red tile in the source image. The visual damage is reduced, but still present.

The right arrow points to an issue with the far right green tile, which has a thin gray line bleeding in from the gray tile in the source image, which is to the left of the green tile.

Tile Map With Image Smoothing Disabled

The two images above were screen captured from Edge. Chrome and Firefox do a better job of hiding the bleeding. Edge seems to bleed on all sides, but Chrome and Firefox seem to only bleed on the right and bottom sides of the source rectangle.

If anyone knows how to fix this please let me know. People ask about this problem in a lot of forums and get work around answers like:

  • Pad your source tiles with border color so the sampler blends in the same color along the edges.
  • Put your source tiles in individual files so the sampler has nothing to sample past the borders.
  • Draw everything to an unscaled buffer canvas and then scale the buffer, ensuring that the sampler is blending in colors from adjacent tiles that are part of the final image, mitigating the visual damage.
  • Draw everything to the unscaled canvas and then scale it using CSS using image-rendering:pixelated, which basically works the same as the previous work around.

I would like to avoid work arounds, however if you know of another one, please post it. I want to know if there is a way to turn off sampling or interpolation or if there is any other way to stop texture bleeding that isn't one of the work arounds I listed.

Here is a fiddle showcasing the issue: https://jsfiddle.net/0rv1upjf/

You can see the same example on my Github Pages page: https://frankpoth.info/pages/javascript-projects/content/texture-bleeding/texture-bleeding.html

Update:

The problem arose due to floating point numbers being used when plotting pixels. The solution is to avoid floats and only draw on integers. Unfortunately, this means setTransform cannot be used efficiently because scaling generally results in floats, but I still managed to keep a good bit of math out of the tile rendering loop. Here's the code:

function drawRounded(source_image, context, scale) {

  var offset_x = -OFFSET.x * scale + context.canvas.width  * 0.5;
  var offset_y = -OFFSET.y * scale + context.canvas.height * 0.5;

  var map_height = (MAP_HEIGHT * scale)|0; // Similar to Math.trunc(MAP_HEIGHT * scale);
  var map_width  = (MAP_WIDTH  * scale)|0;
  var tile_size  = TILE_SIZE * scale;

  var rendered_tile_size = (tile_size + 1)|0; // Similar to Math.ceil(tile_size);

  var map_index = 0; // Track the tile index in the map. This increases once per draw loop.

  /* Loop through all tile positions in actual coordinate space so no additional calculations based on grid index are needed. */
  for (var y = 0; y < map_height; y += tile_size) { // y first so we draw rows from top to bottom

    for (var x = 0; x < map_width; x += tile_size) {

      var frame = FRAMES[MAP[map_index]]; // The frame is the source location of the tile in the source_image.

      // We have to keep the dx, dy truncation inside the loop to ensure the highest level of accuracy possible. 
      context.drawImage(source_image, frame.x, frame.y, TILE_SIZE, TILE_SIZE, (offset_x + x)|0, (offset_y + y)|0, rendered_tile_size, rendered_tile_size);

      map_index ++;

    }

  }

}

I'm using Bitwise OR or the | operator to do my rounding. Bitwise Or returns a 1 in each bit position for which the corresponding bits of either or both operands are 1s. Bitwise operations will convert a float to an int. Using 0 as the right operand will match all the bits in the left operand and truncate the decimals. The downside to this is it only supports 32 bits, but I doubt I'll ever need more than 32 bits for my tile positions.

For example:

-10.5 | 0 == -10

10.1 | 0 == 10

10.5 | 0 == 10

In binary:

1010 | 0000 == 1010

1
readable-email.org/list/whatwg/topic/… In this link they discuss what appears to be the issue. The sampler is grabbing pixels from outside the source rectangle.Frank
You have a retina or hiRes display. Check devicePixelRatio (when page not zoomed). Values > 1 indicate a retina or hiRes device. If so the blur is the result of the canvas being composited onto the page by the DOM and can not fixed using the CanvasRenderingContext2D API. You need to tell the compositor not to smooth the canvas,. Use CSS rule image-rendering: pixelated; for canvas. Some will advise that you up the resolution of the canvas DONT as this squares or more RAM use and render time.Blindman67
Yeah, I definitely don't want to increase the resolution of the canvas. I tried outputting window.devicePixelRatio and I got 1.5. I changed my laptop's display settings to scale the screen to 100% instead of 150% and then I got an output of 1. Even when DPR is 1, I'm still seeing textures bleed. Here's the modified fiddle: jsfiddle.net/89a36nrt I think this is an issue with the browser's nearest neighbor sampling implementation not clamping to the boundaries of the source rectangle, but rather to the boundaries of the source image. This allows it to pull in the unwanted pixels.Frank
Here's a fiddle with image-rendering:pixelated set on the style of the canvas: jsfiddle.net/89a36nrt/1 Unfortunately, I still get the same texture bleeding.By the way, I recognize your handle. Your answers on similar topics have helped me a lot with transforms.Frank
Yeah, it is a good bit of code, but I do appreciate your thoughts thus far in any case. I will try getting rid of floating points wherever possible. I imagine it will help, but unfortunately when it comes to scrolling and scaling at the same time, any rounding to the nearest int causes quite a bit of jitter. I have a fear that this is something vendors have simply overlooked and that the solution is to use WebGL. If that's the case, I will learn WebGL. I mainly wanted to post this question to have a good unified resource on the web that clearly identifies the problem and provides work-arounds.Frank

1 Answers

2
votes

This is a rounding issue.

There was already that question about this issue experienced on Safari browser when the context is translated to exactly n.5, Edge an IE are even worse and always bleed one way or an other, Chrome for macOs bleeds on n.5 too, but only when drawing an <img>, <canvas> are fine.

Least to say, that's a buggy area.

I didn't check the specs to know exactly what they should do, but there is an easy workaround.

Compute yourself the transformation of your coordinates so you can control exactly how they'll get rounded.
This way you don't even need to switch off the image smoothing algorithm, you'll always have sharp pixels since you'll always draw on pixel boundaries:

// First calculate the scaled translations
const scaled_offset_left = -OFFSET.x * scale + context.canvas.width * 0.5;
const scaled_offset_top  = -OFFSET.y * scale + context.canvas.height * 0.5;

// when drawing each tile

const dest_x = Math.floor( scaled_offset_left + (x * scale) );
const dest_y = Math.floor( scaled_offset_top  + (y * scale) );
const dest_size = Math.ceil( TILE_SIZE * scale );

context.drawImage( source_image,
  frame.x, frame.y, TILE_SIZE, TILE_SIZE,
  dest_x, dest_y, dest_size, dest_size,
);

/* This is the tile map. Each value is a frame index in the FRAMES array. Each frame tells drawImage where to blit the source from */
const MAP = [
  0, 0, 0, 1, 1, 1, 1, 2, 2, 2,
  0, 1, 0, 1, 2, 2, 1, 2, 3, 2,
  0, 0, 0, 1, 1, 1, 1, 2, 2, 2,
  3, 3, 3, 4, 4, 4, 4, 5, 5, 5,
  3, 4, 3, 4, 5, 5, 4, 5, 6, 5,
  3, 4, 3, 4, 5, 5, 4, 5, 6, 5,
  3, 3, 3, 4, 4, 4, 4, 5, 5, 5,
  6, 6, 6, 7, 7, 7, 7, 8, 8, 8,
  6, 7, 6, 7, 8, 8, 7, 8, 0, 8,
  6, 6, 6, 7, 7, 7, 7, 8, 8, 8
];

const TILE_SIZE = 8; // Each tile is 8x8 pixels

const MAP_HEIGHT = 80; // The map is 80 pixels tall by 80 pixels wide
const MAP_WIDTH = 80;

/* Each frame represents the source x, y coordinates of a tile in the source image. They are indexed according to the map values */
const FRAMES = [
  { x:0,  y:0 }, // map value = 0
  { x:8,  y:0 }, // map value = 1
  { x:16, y:0 }, // map value = 2
  { x:0,  y:8 }, // etc.
  { x:8,  y:8 },
  { x:16, y:8},
  { x:0,  y:16},
  { x:8,  y:16},
  { x:16, y:16}
];

/* These represent the state of the keyboard keys being used. false is up and true is down */
const KEYS = {
  down: false,
  left: false,
  right: false,
  scale_down: false, // the D key
  scale_up: false, // the F key
  up: false
}

/* This is the scroll offset. You can also think of it as the position of the red dot in the map. */
const OFFSET = {
  x: MAP_WIDTH * 0.5,
  y: MAP_HEIGHT * 0.5
}; // It starts out centered in the map.

const MAX_SCALE = 75; // Max scale is 75 times larger than the actual image size.
const MIN_SCALE = 0; // Texture bleeding seems to only occur on upscale, but min scale is 0 in case you want to try it.

var scale = 4.71; // some arbitrary number that will hopefully cause the issue in your browser

/* Get the canvas drawing context. */
var context = document.querySelector('canvas').getContext('2d', {
  alpha: false,
  desynchronized: true
});

/* The toggle button is the div */
var toggle = document.querySelector('div');

/* The source image is a 24x24 square with 9 tile images of various colors in it. */
var base_64_image_source = 'data:image/png;base64,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';

var source_image = new Image(); // This will be the source image

/* The keyboard event handler */
function keyDownUp(event) {

  var state = event.type == 'keydown' ? true : false;

  switch (event.keyCode) {

    case 37:
      KEYS.left = state;
      break;
    case 38:
      KEYS.up = state;
      break;
    case 39:
      KEYS.right = state;
      break;
    case 40:
      KEYS.down = state;
      break;
    case 68:
      KEYS.scale_down = state;
      break;
    case 70:
      KEYS.scale_up = state;

  }

}

/* This is the update and rendering loop. It handles input and draws the images. */
function loop() {

  window.requestAnimationFrame(loop); // Perpetuate the loop

  /* Prepare to move and scale the image with the keyboard input */
  if (KEYS.left) OFFSET.x -= 0.5;
  if (KEYS.right) OFFSET.x += 0.5;
  if (KEYS.up) OFFSET.y -= 0.5;
  if (KEYS.down) OFFSET.y += 0.5;
  if (KEYS.scale_down) scale -= 0.5 * scale / MAX_SCALE;
  if (KEYS.scale_up) scale += 0.5 * scale / MAX_SCALE;

  /* Keep the scale size within a defined range */
  if (scale > MAX_SCALE) scale = MAX_SCALE;
  else if (scale < MIN_SCALE) scale = MIN_SCALE;

  /* Clear the canvas to gray. */
  context.setTransform(1, 0, 0, 1, 0, 0); // Set the transform back to the identity matrix
  context.fillStyle = "#202830"; // Set the fill color to gray
  context.fillRect(0, 0, context.canvas.width, context.canvas.height); // fill the entire canvas


  /* [EDIT] 
    Don't set the transform, we will calculate it ourselves
    // context.setTransform(scale, 0, 0, scale, -OFFSET.x * scale + context.canvas.width * 0.5, -OFFSET.y * scale + context.canvas.height * 0.5); 
  
    First step is calculating the scaled translation
  */
  
  const scaled_offset_left = -OFFSET.x * scale + context.canvas.width * 0.5;
  const scaled_offset_top  = -OFFSET.y * scale + context.canvas.height * 0.5;

  let map_index = 0; // Track the tile index in the map. This increases once per draw loop.


 
  /* Loop through all tile positions in actual coordinate space so no additional calculations based on grid index are needed. */
  for (let y = 0; y < MAP_HEIGHT; y += TILE_SIZE) { // y first so we draw rows from top to bottom

    for (let x = 0; x < MAP_WIDTH; x += TILE_SIZE) {

      const frame = FRAMES[MAP[map_index]]; // The frame is the source location of the tile in the source_image.

      /* [EDIT] 
        We transform the coordinates ourselves
        We can control a uniform rounding by using floor and ceil
      */

      const dest_x = Math.floor( scaled_offset_left + (x * scale) );
      const dest_y = Math.floor( scaled_offset_top  + (y * scale) );
      const dest_size = Math.ceil(TILE_SIZE * scale);

      context.drawImage( source_image,
        frame.x, frame.y, TILE_SIZE, TILE_SIZE,
        dest_x, dest_y, dest_size, dest_size
      );

      map_index++;

    }

  }

  /* Draw the red dot in the center of the screen. */
  context.fillStyle = "#ff0000";

  /* [EDIT]
    Do the same kind of calculations for the "dot" if you don't want antialiasing

  //  const dot_x = Math.floor( scaled_offset_left + ((OFFSET.x - 0.5) * scale) );
  //  const dot_y = Math.floor( scaled_offset_top + ((OFFSET.y - 0.5) * scale) );
  //  const dot_size = Math.ceil( scale );
  //  context.fillRect( dot_x, dot_y, dot_size, dot_size ); // center on the dot

  But if you do want antialiasing for the dot, then just set the transformation for this drawing
  */
   context.setTransform(scale, 0, 0, scale, scaled_offset_left, scaled_offset_top);
   context.fillRect( (OFFSET.x - 0.5), (OFFSET.y - 0.5), 1, 1 ); // center on the dot

  

  var smoothing = context.imageSmoothingEnabled; // Get the current smoothing value because we are going to ignore it briefly.

  /* Draw the source image in the top left corner for reference. */
  context.setTransform(4, 0, 0, 4, 0, 0); // Zoom in on it so it's visible. 
  context.imageSmoothingEnabled = false; // Set smoothing to false so we get a crisp source image representation (the real source image is not scaled at all).
  context.drawImage( source_image, 0, 0 );
  context.imageSmoothingEnabled = smoothing; // Set smoothing back the way it was according to the toggle choice.

}

/* Turn image smoothing on and off when you press the toggle. */
function toggleSmoothing(event) {

  context.imageSmoothingEnabled = !context.imageSmoothingEnabled;

  if (context.imageSmoothingEnabled) toggle.innerText = 'Smoothing Enabled'; // Make sure the button has appropriate text in it.
  else toggle.innerText = 'Smoothing Disabled';

}

/* The main loop will start after the source image is loaded to ensure there is something to draw. */
source_image.addEventListener('load', (event) => {

  window.requestAnimationFrame(loop); // Start the loop

}, { once: true });

/* Add the toggle smoothing click handler to the div. */
toggle.addEventListener('click', toggleSmoothing);

/* Add keyboard input */
window.addEventListener('keydown', keyDownUp);
window.addEventListener('keyup', keyDownUp);

/* Resize the canvas. */
context.canvas.width = 480;
context.canvas.height = 480;

toggleSmoothing(); // Set imageSmoothingEnabled

/* Load the source image from the base64 string. */
source_image.setAttribute('src', base_64_image_source);
* {
  box-sizing: border-box;
  margin: 0;
  overflow: hidden;
  padding: 0;
  user-select: none;
}

body,
html {
  background-color: #202830;
  color: #ffffff;
  height: 100%;
  width: 100%;
}

body {
  align-items: center;
  display: grid;
  justify-items: center;
}

p {
  max-width: 640px;
}

div {
  border: #ffffff 2px solid;
  bottom: 4px;
  cursor: pointer;
  padding: 8px;
  position: fixed;
  right: 4px
}
<div>Smoothing Disabled</div>
<p>Use the arrow keys to scroll and the D and F keys to scale. The source image is represented on the top left. Notice the vertical and horizontal lines that appear between tiles as you scroll and scale. They are the color of the tile's neighbor in the source
  image. This may be due to color sampling that occurs during scaling. Click the toggle to set imageSmoothingEnabled on the drawing context.</p>

<canvas></canvas>

Note that to draw your "player" dot, you can either choose to do the same caulcations manually to avoid the blurring caused by antialiasing, or if you actually want that blurring, then you can simply set the transform only for this dot. In your position I would probably even make something modular like after a certain scale round, and below smoothen, but I'll let the reader do that implementation.