namespace SixLabors.ImageSharp.Tests.TestUtilities.ImageComparison { using System; using System.Collections.Generic; using System.Runtime.CompilerServices; using SixLabors.ImageSharp.Advanced; using SixLabors.ImageSharp.PixelFormats; using SixLabors.Primitives; public class TolerantImageComparer : ImageComparer { public const float DefaultImageThreshold = 1.0f / (100 * 100 * 255); public TolerantImageComparer(float imageThreshold, int perPixelManhattanThreshold = 0) { this.ImageThreshold = imageThreshold; this.PerPixelManhattanThreshold = perPixelManhattanThreshold; } /// /// The maximal tolerated difference represented by a value between 0.0 and 1.0. /// Examples of percentage differences on a single pixel: /// 1. PixelA = (255,255,255,0) PixelB =(0,0,0,255) leads to 100% difference on a single pixel /// 2. PixelA = (255,255,255,0) PixelB =(255,255,255,255) leads to 25% difference on a single pixel /// 3. PixelA = (255,255,255,0) PixelB =(128,128,128,128) leads to 50% difference on a single pixel /// /// The total differences is the sum of all pixel differences normalized by image dimensions! /// The individual distances are calculated using the Manhattan function: /// /// https://en.wikipedia.org/wiki/Taxicab_geometry /// /// ImageThresholdInPercents = 1.0/255 means that we allow one byte difference per channel on a 1x1 image /// ImageThresholdInPercents = 1.0/(100*100*255) means that we allow only one byte difference per channel on a 100x100 image /// public float ImageThreshold { get; } /// /// The threshold of the individual pixels before they acumulate towards the overall difference. /// For an individual pixel pair the value is the Manhattan distance of pixels: /// /// https://en.wikipedia.org/wiki/Taxicab_geometry /// /// public int PerPixelManhattanThreshold { get; } public override ImageSimilarityReport CompareImagesOrFrames(ImageFrame expected, ImageFrame actual) { if (expected.Size() != actual.Size()) { throw new InvalidOperationException("Calling ImageComparer is invalid when dimensions mismatch!"); } int width = actual.Width; // TODO: Comparing through Rgba32 is not robust enough because of the existance of super high precision pixel types. Rgba32[] aBuffer = new Rgba32[width]; Rgba32[] bBuffer = new Rgba32[width]; float totalDifference = 0.0f; var differences = new List(); for (int y = 0; y < actual.Height; y++) { Span aSpan = expected.GetPixelRowSpan(y); Span bSpan = actual.GetPixelRowSpan(y); PixelOperations.Instance.ToRgba32(aSpan, aBuffer, width); PixelOperations.Instance.ToRgba32(bSpan, bBuffer, width); for (int x = 0; x < width; x++) { int d = GetManhattanDistanceInRgbaSpace(ref aBuffer[x], ref bBuffer[x]); if (d > this.PerPixelManhattanThreshold) { var diff = new PixelDifference(new Point(x, y), aBuffer[x], bBuffer[x]); differences.Add(diff); totalDifference += d; } } } float normalizedDifference = totalDifference / ((float)actual.Width * (float)actual.Height); normalizedDifference /= 4.0f * 255.0f; if (normalizedDifference > this.ImageThreshold) { return new ImageSimilarityReport(expected, actual, differences, normalizedDifference); } else { return ImageSimilarityReport.Empty; } } [MethodImpl(MethodImplOptions.AggressiveInlining)] private static int GetManhattanDistanceInRgbaSpace(ref Rgba32 a, ref Rgba32 b) { return Diff(a.R, b.R) + Diff(a.G, b.G) + Diff(a.B, b.B) + Diff(a.A, b.A); } [MethodImpl(MethodImplOptions.AggressiveInlining)] private static int Diff(byte a, byte b) => Math.Abs(a - b); } }