/// <summary>
/// Returns contours that are relevant across two input images
/// </summary>
/// <param name="img1">First image</param>
/// <param name="img2">Second image</param>
/// <param name="differenceThreshold">How much difference is considered relevant before we consider a pixel as a contour element (value between 0,255)</param>
/// <param name="areaThreshold">How big should be the area of the contour, before it is considered relevant (min value 2)</param>
/// <param name="kernelSize">Gaussian blur kernel size (i.e. 11)</param>
/// <param name="sigma">Gaussian kernel sigma over X and Y directions (i.e. 1.5)</param>
/// <param name="borderWidth">Allowed border width (i.e. 5)</param>
/// <returns>List of contours, thresholded and SSIM image</returns>
public SSIM FindContours(GrayImage img1, GrayImage img2,
int differenceThreshold,
int areaThreshold,
int kernelSize,
double sigma,
int borderWidth)
{
if (img1.Width != img2.Width)
{
throw new ArgumentException(nameof(img1.Width));
}
if (differenceThreshold < 0 || differenceThreshold > 255)
{
throw new AggregateException(nameof(differenceThreshold));
}
if (areaThreshold < 2)
{
throw new ArgumentException(nameof(areaThreshold));
}
double[,] kernel2d = GaussianBlurKernel.Kernel2D(kernelSize, sigma);
var ux = img1.GaussianBlur(kernel2d);
var uy = img2.GaussianBlur(kernel2d);
var uxx = img1.Multiply(img1).GaussianBlur(kernel2d);
var uyy = img2.Multiply(img2).GaussianBlur(kernel2d);
var uxy = img1.Multiply(img2).GaussianBlur(kernel2d);
var uxSquared = ux.Multiply(ux);
var vx = uxx.Subtract(uxSquared);
var uySquared = uy.Multiply(uy);
var vy = uyy.Subtract(uySquared);
var uxUy = ux.Multiply(uy);
var vxy = uxy.Subtract(uxUy);
double c1 = Math.Pow(K1 * L, 2);
double c2 = Math.Pow(K2 * L, 2);
var a1 = uxUy.Convert(x => (float)(2 * x + c1)); //2 * ux * uy + C1,
var a2 = vxy.Convert(x => (float)(2 * x + c2)); //2 * vxy + C2
var b1 = uxSquared.Add(uySquared).Convert(x => (float)(x + c1)); //ux ** 2 + uy ** 2 + C1
var b2 = vx.Add(vy).Convert(x => (float)(x + c2)); //vx + vy + C2)
var d = b1.Multiply(b2);
var s = a1.Multiply(a2).Divide(d);
byte[][] ssimImage = s.ConvertAndUnwrap(x => (byte)(x * byte.MaxValue));
byte[][] thresholdInvImage = ssimImage
.Skip(borderWidth)
.Take(img1.Height - 2 * borderWidth)
.Select(x => x.Skip(borderWidth).Take(img1.Width - 2 * borderWidth).ToArray())
.ToArray();
ImageService.ThresholdInvInPlace(thresholdInvImage, differenceThreshold, byte.MaxValue);
return(new SSIM(ssimImage, thresholdInvImage, Contour.FindContours(thresholdInvImage, byte.MaxValue, areaThreshold)));
}
public GrayImage GaussianBlur(int size, double sigma)
{
var kernel = GaussianBlurKernel.Kernel2D(size, sigma);
return(new GrayImage(Convolve(image, kernel)));
}