image_pad_or_crop

image_pad_or_crop.js

function image_pad_or_crop(image, output_height, output_width) {
  // Pads or crops an image so that the height and width dimensions match output_height and output_width. 
  // This function is compatible with tf.image.resize_with_crop_or_pad(image, target_height, target_width)

  // Validate inputs
  const image_size = image.height * image.width * image.depth;
  if (image.data.length != image_size) {
    console.log("Error: image dimensions do not match rgba buffer size!");
    return null;
  }
  if (image.data.constructor !== Uint8ClampedArray && image.data.constructor !== Float32Array) {
    console.log("Error: image data should be of type Uint8ClampedArray or Float32Array!");
    return null;
  }
  if (output_height <= 0 || output_width <= 0)
  {
    console.log("Error: output_height and output_width should be positive integers ( > 0)!");
    return null;
  }

  // Compute output dimensions
  const output_depth = image.depth;

  // Init output image        
  const output = {
    height: output_height,
    width: output_width,
    depth: output_depth,
    data: new image.data.constructor(output_height * output_width * output_depth) // Initialized with zeros
  }

  // Compute padding/cropping amounts
  function floor(v) { return (v >= 0 || -1) * Math.floor(Math.abs(v));}  // use custom floor that is compatible with tf.image.resize_with_crop_or_pad
  const top = floor((output_height - image.height) / 2);
  const bottom = output_height - image.height - top;
  const left = floor((output_width - image.width) / 2);
  const right = output_width - image.width - left;

  // Copy input image into output  
  let output_idx = Math.max(top, 0) * output.width * output.depth + Math.max(left, 0) * output_depth;
  let image_idx = Math.max(-top, 0) * image.width * image.depth + Math.max(-left, 0) * image.depth;  
  for (let y = 0; y < (output_height - Math.max(top, 0) - Math.max(bottom, 0)); y++, image_idx += image.width * image.depth, output_idx += output_width * output.depth)
    output.data.set(image.data.slice(image_idx, image_idx +  (image.width - Math.max(-left, 0) - Math.max(-right, 0)) * image.depth), output_idx)

  return output;
}

// Utility function (only for testing)
function show_image(image) {
  str = "";
  let idx = 0;
  for (let y = 0; y < image.height; y++) {
    for (let x = 0; x < image.width; x++) {
      str += "[ ";
      for (let z = 0; z < image.depth; z++)
        str += image.data[idx++] + " ";
      str += "] ";
    }
    str += "\n";
  }

  console.log(str);
}

// Test
test_image = {
  "height": 3,
  "width": 4,
  "depth": 1,
  "data": (new Float32Array([...Array(3*4*1).keys()]))
}
show_image(test_image);
/* should output:
[ 0 ] [ 1 ] [ 2 ] [ 3 ] 
[ 4 ] [ 5 ] [ 6 ] [ 7 ] 
[ 8 ] [ 9 ] [ 10 ] [ 11 ] 
*/
output = image_pad_or_crop(test_image, 6, 6);
show_image(output);
/* should output:
[ 0 ] [ 0 ] [ 0 ] [ 0 ] [ 0 ] [ 0 ] 
[ 0 ] [ 0 ] [ 1 ] [ 2 ] [ 3 ] [ 0 ] 
[ 0 ] [ 4 ] [ 5 ] [ 6 ] [ 7 ] [ 0 ] 
[ 0 ] [ 8 ] [ 9 ] [ 10 ] [ 11 ] [ 0 ] 
[ 0 ] [ 0 ] [ 0 ] [ 0 ] [ 0 ] [ 0 ] 
[ 0 ] [ 0 ] [ 0 ] [ 0 ] [ 0 ] [ 0 ] 
*/