Any difference?

[coenm]

Any difference?

[LambertWM]

Interesting !
In what way are the results from pHash better than ImageHash?

[developervariety]

Thank you for the response.

I use pHash to compare similar images for a a service I try to automate. With the result of pHash I take the image and submit it to the service. For some reason, the results from this library are almost always incorrect.

It confuses me on why exactly they are incorrect if they are both based off the same library. I will provide some data so you can test for yourself, shortly.

[developervariety]

Maybe @pgrho could look into this whenever they have free time on their hands.

More context, I compare images with all colors made black. The only white part is the transparency if that makes any sense.

[pgrho]

@developervariety My pHash creates an intermediate image of 32x32 pixels to calculate DCT, but 64x64 in the ImageHash. So they seems to be incompatible.

ImageHash/src/ImageHash/HashAlgorithms/PerceptualHash.cs

Lines 18 to 81 in ae66ced

	private const int Size = 64;
	private static readonly double Sqrt2DivSize = Math.Sqrt(2D / Size);
	private static readonly double Sqrt2 = 1 / Math.Sqrt(2);
	/// <inheritdoc />
	public ulong Hash(Image<Rgba32> image)
	{
	if (image == null)
	throw new ArgumentNullException(nameof(image));
	var rows = new double[Size][];
	var sequence = new double[Size];
	var matrix = new double[Size][];
	image.Mutate(ctx => ctx
	.Resize(Size, Size)
	.Grayscale(GrayscaleMode.Bt601)
	.AutoOrient());
	// Calculate the DCT for each row.
	for (var y = 0; y < Size; y++)
	{
	for (var x = 0; x < Size; x++)
	sequence[x] = image[x, y].R;
	rows[y] = Dct1D(sequence);
	}
	// Calculate the DCT for each column.
	for (var x = 0; x < Size; x++)
	{
	for (var y = 0; y < Size; y++)
	sequence[y] = rows[y][x];
	matrix[x] = Dct1D(sequence);
	}
	// Only use the top 8x8 values.
	var top8X8 = new List<double>(Size);
	for (var y = 0; y < 8; y++)
	{
	for (var x = 0; x < 8; x++)
	top8X8.Add(matrix[y][x]);
	}
	var topRight = top8X8.ToArray();
	// Get Median.
	var median = CalculateMedian64Values(topRight);
	// Calculate hash.
	var mask = 1UL << (Size - 1);
	var hash = 0UL;
	for (var i = 0; i < Size; i++)
	{
	if (topRight[i] > median)
	hash |= mask;
	mask = mask >> 1;
	}
	return hash;
	}

https://github.com/pgrho/phash/blob/ab46a05226961750a59003494c1a95a4ae8b0131/src/Shipwreck.Phash/ImagePhash.cs#L167-L196

[developervariety]

Thank you for looking into this @pgrho. Very much appreciated, I'll stick to what works for my project for now.