diff --git a/src/ImageSharp/Advanced/AotCompilerTools.cs b/src/ImageSharp/Advanced/AotCompilerTools.cs
index 9e7624480d..eb6991e6a1 100644
--- a/src/ImageSharp/Advanced/AotCompilerTools.cs
+++ b/src/ImageSharp/Advanced/AotCompilerTools.cs
@@ -1,6 +1,9 @@
 // Copyright (c) Six Labors and contributors.
 // Licensed under the Apache License, Version 2.0.
 
+using System.Numerics;
+using SixLabors.ImageSharp.Formats;
+using SixLabors.ImageSharp.Formats.Jpeg.Components;
 using SixLabors.ImageSharp.PixelFormats;
 using SixLabors.ImageSharp.Processing.Processors.Dithering;
 using SixLabors.ImageSharp.Processing.Processors.Quantization;
@@ -15,6 +18,17 @@ namespace SixLabors.ImageSharp.Advanced
     /// </summary>
     public static class AotCompilerTools
     {
+        static AotCompilerTools()
+        {
+            System.Runtime.CompilerServices.Unsafe.SizeOf<long>();
+            System.Runtime.CompilerServices.Unsafe.SizeOf<short>();
+            System.Runtime.CompilerServices.Unsafe.SizeOf<float>();
+            System.Runtime.CompilerServices.Unsafe.SizeOf<double>();
+            System.Runtime.CompilerServices.Unsafe.SizeOf<byte>();
+            System.Runtime.CompilerServices.Unsafe.SizeOf<Block8x8>();
+            System.Runtime.CompilerServices.Unsafe.SizeOf<Vector4>();
+        }
+
         /// <summary>
         /// Seeds the compiler using the given pixel format.
         /// </summary>
@@ -27,6 +41,13 @@ namespace SixLabors.ImageSharp.Advanced
             AotCompileWuQuantizer<TPixel>();
             AotCompileDithering<TPixel>();
 
+            System.Runtime.CompilerServices.Unsafe.SizeOf<TPixel>();
+
+            AotCodec<TPixel>(new Formats.Png.PngDecoder(), new Formats.Png.PngEncoder());
+            AotCodec<TPixel>(new Formats.Bmp.BmpDecoder(), new Formats.Bmp.BmpEncoder());
+            AotCodec<TPixel>(new Formats.Gif.GifDecoder(), new Formats.Gif.GifEncoder());
+            AotCodec<TPixel>(new Formats.Jpeg.JpegDecoder(), new Formats.Jpeg.JpegEncoder());
+
             // TODO: Do the discovery work to figure out what works and what doesn't.
         }
 
@@ -99,5 +120,31 @@ namespace SixLabors.ImageSharp.Advanced
             TPixel pixel = default;
             test.Dither<TPixel>(new ImageFrame<TPixel>(Configuration.Default, 1, 1), pixel, pixel, 0, 0, 0, 0, 0, 0);
         }
+
+        /// <summary>
+        /// This method pre-seeds the decoder and encoder for a given pixel format in the AoT compiler for iOS.
+        /// </summary>
+        /// <param name="decoder">The image decoder to seed.</param>
+        /// <param name="encoder">The image encoder to seed.</param>
+        /// <typeparam name="TPixel">The pixel format.</typeparam>
+        private static void AotCodec<TPixel>(IImageDecoder decoder, IImageEncoder encoder)
+            where TPixel : struct, IPixel<TPixel>
+        {
+            try
+            {
+                decoder.Decode<TPixel>(Configuration.Default, null);
+            }
+            catch
+            {
+            }
+
+            try
+            {
+                encoder.Encode<TPixel>(null, null);
+            }
+            catch
+            {
+            }
+        }
     }
 }
\ No newline at end of file
diff --git a/src/ImageSharp/Formats/Bmp/BmpDecoderCore.cs b/src/ImageSharp/Formats/Bmp/BmpDecoderCore.cs
index ef3ca24ee8..68528edcd6 100644
--- a/src/ImageSharp/Formats/Bmp/BmpDecoderCore.cs
+++ b/src/ImageSharp/Formats/Bmp/BmpDecoderCore.cs
@@ -108,7 +108,7 @@ namespace SixLabors.ImageSharp.Formats.Bmp
         {
             try
             {
-                this.ReadImageHeaders(stream, out bool inverted, out byte[] palette);
+                int bytesPerColorMapEntry = this.ReadImageHeaders(stream, out bool inverted, out byte[] palette);
 
                 var image = new Image<TPixel>(this.configuration, this.infoHeader.Width, this.infoHeader.Height, this.metaData);
 
@@ -137,6 +137,7 @@ namespace SixLabors.ImageSharp.Formats.Bmp
                                 this.infoHeader.Width,
                                 this.infoHeader.Height,
                                 this.infoHeader.BitsPerPixel,
+                                bytesPerColorMapEntry,
                                 inverted);
                         }
 
@@ -329,18 +330,20 @@ namespace SixLabors.ImageSharp.Formats.Bmp
         /// <param name="colors">The <see cref="T:byte[]"/> containing the colors.</param>
         /// <param name="width">The width of the bitmap.</param>
         /// <param name="height">The height of the bitmap.</param>
-        /// <param name="bits">The number of bits per pixel.</param>
+        /// <param name="bitsPerPixel">The number of bits per pixel.</param>
+        /// <param name="bytesPerColorMapEntry">Usually 4 bytes, but in case of Windows 2.x bitmaps or OS/2 1.x bitmaps
+        /// the bytes per color palette entry's can be 3 bytes instead of 4.</param>
         /// <param name="inverted">Whether the bitmap is inverted.</param>
-        private void ReadRgbPalette<TPixel>(Buffer2D<TPixel> pixels, byte[] colors, int width, int height, int bits, bool inverted)
+        private void ReadRgbPalette<TPixel>(Buffer2D<TPixel> pixels, byte[] colors, int width, int height, int bitsPerPixel, int bytesPerColorMapEntry, bool inverted)
             where TPixel : struct, IPixel<TPixel>
         {
             // Pixels per byte (bits per pixel)
-            int ppb = 8 / bits;
+            int ppb = 8 / bitsPerPixel;
 
             int arrayWidth = (width + ppb - 1) / ppb;
 
             // Bit mask
-            int mask = 0xFF >> (8 - bits);
+            int mask = 0xFF >> (8 - bitsPerPixel);
 
             // Rows are aligned on 4 byte boundaries
             int padding = arrayWidth % 4;
@@ -366,7 +369,7 @@ namespace SixLabors.ImageSharp.Formats.Bmp
                         int colOffset = x * ppb;
                         for (int shift = 0, newX = colOffset; shift < ppb && newX < width; shift++, newX++)
                         {
-                            int colorIndex = ((rowSpan[offset] >> (8 - bits - (shift * bits))) & mask) * 4;
+                            int colorIndex = ((rowSpan[offset] >> (8 - bitsPerPixel - (shift * bitsPerPixel))) & mask) * bytesPerColorMapEntry;
 
                             color.FromBgr24(Unsafe.As<byte, Bgr24>(ref colors[colorIndex]));
                             pixelRow[newX] = color;
@@ -510,6 +513,11 @@ namespace SixLabors.ImageSharp.Formats.Bmp
                 // 12 bytes
                 this.infoHeader = BmpInfoHeader.ParseCore(buffer);
             }
+            else if (headerSize == BmpInfoHeader.Os22ShortSize)
+            {
+                // 16 bytes
+                this.infoHeader = BmpInfoHeader.ParseOs22Short(buffer);
+            }
             else if (headerSize >= BmpInfoHeader.Size)
             {
                 // >= 40 bytes
@@ -571,7 +579,9 @@ namespace SixLabors.ImageSharp.Formats.Bmp
         /// <summary>
         /// Reads the <see cref="BmpFileHeader"/> and <see cref="BmpInfoHeader"/> from the stream and sets the corresponding fields.
         /// </summary>
-        private void ReadImageHeaders(Stream stream, out bool inverted, out byte[] palette)
+        /// <returns>Bytes per color palette entry. Usually 4 bytes, but in case of Windows 2.x bitmaps or OS/2 1.x bitmaps
+        /// the bytes per color palette entry's can be 3 bytes instead of 4.</returns>
+        private int ReadImageHeaders(Stream stream, out bool inverted, out byte[] palette)
         {
             this.stream = stream;
 
@@ -591,6 +601,7 @@ namespace SixLabors.ImageSharp.Formats.Bmp
             }
 
             int colorMapSize = -1;
+            int bytesPerColorMapEntry = 4;
 
             if (this.infoHeader.ClrUsed == 0)
             {
@@ -598,12 +609,15 @@ namespace SixLabors.ImageSharp.Formats.Bmp
                     || this.infoHeader.BitsPerPixel == 4
                     || this.infoHeader.BitsPerPixel == 8)
                 {
-                    colorMapSize = ImageMaths.GetColorCountForBitDepth(this.infoHeader.BitsPerPixel) * 4;
+                    int colorMapSizeBytes = this.fileHeader.Offset - BmpFileHeader.Size - this.infoHeader.HeaderSize;
+                    int colorCountForBitDepth = ImageMaths.GetColorCountForBitDepth(this.infoHeader.BitsPerPixel);
+                    bytesPerColorMapEntry = colorMapSizeBytes / colorCountForBitDepth;
+                    colorMapSize = colorMapSizeBytes;
                 }
             }
             else
             {
-                colorMapSize = this.infoHeader.ClrUsed * 4;
+                colorMapSize = this.infoHeader.ClrUsed * bytesPerColorMapEntry;
             }
 
             palette = null;
@@ -622,6 +636,8 @@ namespace SixLabors.ImageSharp.Formats.Bmp
             }
 
             this.infoHeader.VerifyDimensions();
+
+            return bytesPerColorMapEntry;
         }
     }
 }
\ No newline at end of file
diff --git a/src/ImageSharp/Formats/Bmp/BmpInfoHeader.cs b/src/ImageSharp/Formats/Bmp/BmpInfoHeader.cs
index 4dd63a9626..5177bc325c 100644
--- a/src/ImageSharp/Formats/Bmp/BmpInfoHeader.cs
+++ b/src/ImageSharp/Formats/Bmp/BmpInfoHeader.cs
@@ -26,6 +26,11 @@ namespace SixLabors.ImageSharp.Formats.Bmp
         /// </summary>
         public const int CoreSize = 12;
 
+        /// <summary>
+        /// Defines the size of the short variant of the OS22XBITMAPHEADER data structure in the bitmap file.
+        /// </summary>
+        public const int Os22ShortSize = 16;
+
         /// <summary>
         /// Defines the size of the biggest supported header data structure in the bitmap file.
         /// </summary>
@@ -143,7 +148,7 @@ namespace SixLabors.ImageSharp.Formats.Bmp
         /// <summary>
         /// Parses the BITMAPCOREHEADER consisting of the headerSize, width, height, planes, and bitsPerPixel fields (12 bytes).
         /// </summary>
-        /// <param name="data">The data to parse,</param>
+        /// <param name="data">The data to parse.</param>
         /// <returns>Parsed header</returns>
         /// <seealso href="https://msdn.microsoft.com/en-us/library/windows/desktop/dd183372.aspx"/>
         public static BmpInfoHeader ParseCore(ReadOnlySpan<byte> data)
@@ -156,6 +161,23 @@ namespace SixLabors.ImageSharp.Formats.Bmp
                 bitsPerPixel: BinaryPrimitives.ReadInt16LittleEndian(data.Slice(10, 2)));
         }
 
+        /// <summary>
+        /// Parses a short variant of the OS22XBITMAPHEADER. It is identical to the BITMAPCOREHEADER, except that the width and height
+        /// are 4 bytes instead of 2.
+        /// </summary>
+        /// <param name="data">The data to parse.</param>
+        /// <returns>Parsed header</returns>
+        /// <seealso href="https://www.fileformat.info/format/os2bmp/egff.htm"/>
+        public static BmpInfoHeader ParseOs22Short(ReadOnlySpan<byte> data)
+        {
+            return new BmpInfoHeader(
+                headerSize: BinaryPrimitives.ReadInt32LittleEndian(data.Slice(0, 4)),
+                width: BinaryPrimitives.ReadInt32LittleEndian(data.Slice(4, 4)),
+                height: BinaryPrimitives.ReadInt32LittleEndian(data.Slice(8, 4)),
+                planes: BinaryPrimitives.ReadInt16LittleEndian(data.Slice(12, 2)),
+                bitsPerPixel: BinaryPrimitives.ReadInt16LittleEndian(data.Slice(14, 2)));
+        }
+
         public unsafe void WriteTo(Span<byte> buffer)
         {
             ref BmpInfoHeader dest = ref Unsafe.As<byte, BmpInfoHeader>(ref MemoryMarshal.GetReference(buffer));
diff --git a/src/ImageSharp/Primitives/DenseMatrix{T}.cs b/src/ImageSharp/Primitives/DenseMatrix{T}.cs
index 7cfa98ec1b..170292e29e 100644
--- a/src/ImageSharp/Primitives/DenseMatrix{T}.cs
+++ b/src/ImageSharp/Primitives/DenseMatrix{T}.cs
@@ -110,7 +110,7 @@ namespace SixLabors.ImageSharp.Primitives
         /// <returns>The <see typeparam="T"/> at the specified position.</returns>
         public ref T this[int row, int column]
         {
-            [MethodImpl(MethodImplOptions.AggressiveInlining)]
+            [MethodImpl(InliningOptions.ShortMethod)]
             get
             {
                 this.CheckCoordinates(row, column);
@@ -125,7 +125,7 @@ namespace SixLabors.ImageSharp.Primitives
         /// <returns>
         /// The <see cref="DenseMatrix{T}"/> representation on the source data.
         /// </returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        [MethodImpl(InliningOptions.ShortMethod)]
         public static implicit operator DenseMatrix<T>(T[,] data) => new DenseMatrix<T>(data);
 
         /// <summary>
@@ -135,9 +135,9 @@ namespace SixLabors.ImageSharp.Primitives
         /// <returns>
         /// The <see cref="T:T[,]"/> representation on the source data.
         /// </returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        [MethodImpl(InliningOptions.ShortMethod)]
 #pragma warning disable SA1008 // Opening parenthesis should be spaced correctly
-        public static implicit operator T[,] (DenseMatrix<T> data)
+        public static implicit operator T[,] (in DenseMatrix<T> data)
 #pragma warning restore SA1008 // Opening parenthesis should be spaced correctly
         {
             var result = new T[data.Rows, data.Columns];
@@ -154,17 +154,38 @@ namespace SixLabors.ImageSharp.Primitives
             return result;
         }
 
+        /// <summary>
+        /// Transposes the rows and columns of the dense matrix.
+        /// </summary>
+        /// <returns>The <see cref="DenseMatrix{T}"/>.</returns>
+        [MethodImpl(InliningOptions.ShortMethod)]
+        public DenseMatrix<T> Transpose()
+        {
+            var result = new DenseMatrix<T>(this.Rows, this.Columns);
+
+            for (int y = 0; y < this.Rows; y++)
+            {
+                for (int x = 0; x < this.Columns; x++)
+                {
+                    ref T value = ref result[x, y];
+                    value = this[y, x];
+                }
+            }
+
+            return result;
+        }
+
         /// <summary>
         /// Fills the matrix with the given value
         /// </summary>
         /// <param name="value">The value to fill each item with</param>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        [MethodImpl(InliningOptions.ShortMethod)]
         public void Fill(T value) => this.Span.Fill(value);
 
         /// <summary>
         /// Clears the matrix setting each value to the default value for the element type
         /// </summary>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        [MethodImpl(InliningOptions.ShortMethod)]
         public void Clear() => this.Span.Clear();
 
         /// <summary>
diff --git a/src/ImageSharp/Processing/Processors/Convolution/BoxBlurProcessor.cs b/src/ImageSharp/Processing/Processors/Convolution/BoxBlurProcessor.cs
index 38dc638b90..644d6c9e17 100644
--- a/src/ImageSharp/Processing/Processors/Convolution/BoxBlurProcessor.cs
+++ b/src/ImageSharp/Processing/Processors/Convolution/BoxBlurProcessor.cs
@@ -29,8 +29,8 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         {
             this.Radius = radius;
             this.kernelSize = (radius * 2) + 1;
-            this.KernelX = this.CreateBoxKernel(true);
-            this.KernelY = this.CreateBoxKernel(false);
+            this.KernelX = this.CreateBoxKernel();
+            this.KernelY = this.KernelX.Transpose();
         }
 
         /// <summary>
@@ -49,24 +49,18 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         public DenseMatrix<float> KernelY { get; }
 
         /// <inheritdoc/>
-        protected override void OnFrameApply(ImageFrame<TPixel> source, Rectangle sourceRectangle, Configuration configuration)
-        {
-            new Convolution2PassProcessor<TPixel>(this.KernelX, this.KernelY).Apply(source, sourceRectangle, configuration);
-        }
+        protected override void OnFrameApply(ImageFrame<TPixel> source, Rectangle sourceRectangle, Configuration configuration) => new Convolution2PassProcessor<TPixel>(this.KernelX, this.KernelY).Apply(source, sourceRectangle, configuration);
 
         /// <summary>
         /// Create a 1 dimensional Box kernel.
         /// </summary>
-        /// <param name="horizontal">Whether to calculate a horizontal kernel.</param>
         /// <returns>The <see cref="DenseMatrix{T}"/></returns>
-        private DenseMatrix<float> CreateBoxKernel(bool horizontal)
+        private DenseMatrix<float> CreateBoxKernel()
         {
             int size = this.kernelSize;
-            DenseMatrix<float> kernel = horizontal
-                ? new DenseMatrix<float>(size, 1)
-                : new DenseMatrix<float>(1, size);
+            var kernel = new DenseMatrix<float>(size, 1);
 
-            kernel.Fill(1.0F / size);
+            kernel.Fill(1F / size);
 
             return kernel;
         }
diff --git a/src/ImageSharp/Processing/Processors/Convolution/GaussianBlurProcessor.cs b/src/ImageSharp/Processing/Processors/Convolution/GaussianBlurProcessor.cs
index 3045b9993f..b3bc15d391 100644
--- a/src/ImageSharp/Processing/Processors/Convolution/GaussianBlurProcessor.cs
+++ b/src/ImageSharp/Processing/Processors/Convolution/GaussianBlurProcessor.cs
@@ -4,7 +4,6 @@
 using System;
 using SixLabors.ImageSharp.PixelFormats;
 using SixLabors.ImageSharp.Primitives;
-using SixLabors.ImageSharp.Processing.Processors;
 using SixLabors.Primitives;
 
 namespace SixLabors.ImageSharp.Processing.Processors.Convolution
@@ -26,11 +25,10 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         /// </summary>
         /// <param name="sigma">The 'sigma' value representing the weight of the blur.</param>
         public GaussianBlurProcessor(float sigma = 3F)
+            : this(sigma, (int)MathF.Ceiling(sigma * 3))
         {
-            this.kernelSize = ((int)Math.Ceiling(sigma) * 2) + 1;
-            this.Sigma = sigma;
-            this.KernelX = this.CreateGaussianKernel(true);
-            this.KernelY = this.CreateGaussianKernel(false);
+            // Kernel radius is calculated using the minimum viable value.
+            // http://chemaguerra.com/gaussian-filter-radius/
         }
 
         /// <summary>
@@ -40,11 +38,8 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         /// The 'radius' value representing the size of the area to sample.
         /// </param>
         public GaussianBlurProcessor(int radius)
+            : this(radius / 3F, radius)
         {
-            this.kernelSize = (radius * 2) + 1;
-            this.Sigma = radius;
-            this.KernelX = this.CreateGaussianKernel(true);
-            this.KernelY = this.CreateGaussianKernel(false);
         }
 
         /// <summary>
@@ -61,8 +56,8 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         {
             this.kernelSize = (radius * 2) + 1;
             this.Sigma = sigma;
-            this.KernelX = this.CreateGaussianKernel(true);
-            this.KernelY = this.CreateGaussianKernel(false);
+            this.KernelX = this.CreateGaussianKernel();
+            this.KernelY = this.KernelX.Transpose();
         }
 
         /// <summary>
@@ -82,22 +77,17 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
 
         /// <inheritdoc/>
         protected override void OnFrameApply(ImageFrame<TPixel> source, Rectangle sourceRectangle, Configuration configuration)
-        {
-            new Convolution2PassProcessor<TPixel>(this.KernelX, this.KernelY).Apply(source, sourceRectangle, configuration);
-        }
+            => new Convolution2PassProcessor<TPixel>(this.KernelX, this.KernelY).Apply(source, sourceRectangle, configuration);
 
         /// <summary>
         /// Create a 1 dimensional Gaussian kernel using the Gaussian G(x) function
         /// </summary>
-        /// <param name="horizontal">Whether to calculate a horizontal kernel.</param>
         /// <returns>The <see cref="DenseMatrix{T}"/></returns>
-        private DenseMatrix<float> CreateGaussianKernel(bool horizontal)
+        private DenseMatrix<float> CreateGaussianKernel()
         {
             int size = this.kernelSize;
             float weight = this.Sigma;
-            DenseMatrix<float> kernel = horizontal
-                ? new DenseMatrix<float>(size, 1)
-                : new DenseMatrix<float>(1, size);
+            var kernel = new DenseMatrix<float>(size, 1);
 
             float sum = 0F;
             float midpoint = (size - 1) / 2F;
@@ -107,30 +97,13 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
                 float x = i - midpoint;
                 float gx = ImageMaths.Gaussian(x, weight);
                 sum += gx;
-                if (horizontal)
-                {
-                    kernel[0, i] = gx;
-                }
-                else
-                {
-                    kernel[i, 0] = gx;
-                }
+                kernel[0, i] = gx;
             }
 
             // Normalize kernel so that the sum of all weights equals 1
-            if (horizontal)
-            {
-                for (int i = 0; i < size; i++)
-                {
-                    kernel[0, i] = kernel[0, i] / sum;
-                }
-            }
-            else
+            for (int i = 0; i < size; i++)
             {
-                for (int i = 0; i < size; i++)
-                {
-                    kernel[i, 0] = kernel[i, 0] / sum;
-                }
+                kernel[0, i] /= sum;
             }
 
             return kernel;
diff --git a/src/ImageSharp/Processing/Processors/Convolution/GaussianSharpenProcessor.cs b/src/ImageSharp/Processing/Processors/Convolution/GaussianSharpenProcessor.cs
index 18963c73c0..786bf7757a 100644
--- a/src/ImageSharp/Processing/Processors/Convolution/GaussianSharpenProcessor.cs
+++ b/src/ImageSharp/Processing/Processors/Convolution/GaussianSharpenProcessor.cs
@@ -27,11 +27,10 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         /// The 'sigma' value representing the weight of the sharpening.
         /// </param>
         public GaussianSharpenProcessor(float sigma = 3F)
+            : this(sigma, (int)MathF.Ceiling(sigma * 3))
         {
-            this.kernelSize = ((int)Math.Ceiling(sigma) * 2) + 1;
-            this.Sigma = sigma;
-            this.KernelX = this.CreateGaussianKernel(true);
-            this.KernelY = this.CreateGaussianKernel(false);
+            // Kernel radius is calculated using the minimum viable value.
+            // http://chemaguerra.com/gaussian-filter-radius/
         }
 
         /// <summary>
@@ -41,11 +40,8 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         /// The 'radius' value representing the size of the area to sample.
         /// </param>
         public GaussianSharpenProcessor(int radius)
+            : this(radius / 3F, radius)
         {
-            this.kernelSize = (radius * 2) + 1;
-            this.Sigma = radius;
-            this.KernelX = this.CreateGaussianKernel(true);
-            this.KernelY = this.CreateGaussianKernel(false);
         }
 
         /// <summary>
@@ -62,8 +58,8 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         {
             this.kernelSize = (radius * 2) + 1;
             this.Sigma = sigma;
-            this.KernelX = this.CreateGaussianKernel(true);
-            this.KernelY = this.CreateGaussianKernel(false);
+            this.KernelX = this.CreateGaussianKernel();
+            this.KernelY = this.KernelX.Transpose();
         }
 
         /// <summary>
@@ -83,91 +79,49 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
 
         /// <inheritdoc/>
         protected override void OnFrameApply(ImageFrame<TPixel> source, Rectangle sourceRectangle, Configuration configuration)
-        {
-            new Convolution2PassProcessor<TPixel>(this.KernelX, this.KernelY).Apply(source, sourceRectangle, configuration);
-        }
+            => new Convolution2PassProcessor<TPixel>(this.KernelX, this.KernelY).Apply(source, sourceRectangle, configuration);
 
         /// <summary>
         /// Create a 1 dimensional Gaussian kernel using the Gaussian G(x) function
         /// </summary>
-        /// <param name="horizontal">Whether to calculate a horizontal kernel.</param>
         /// <returns>The <see cref="DenseMatrix{T}"/></returns>
-        private DenseMatrix<float> CreateGaussianKernel(bool horizontal)
+        private DenseMatrix<float> CreateGaussianKernel()
         {
             int size = this.kernelSize;
             float weight = this.Sigma;
-            DenseMatrix<float> kernel = horizontal
-                ? new DenseMatrix<float>(size, 1)
-                : new DenseMatrix<float>(1, size);
+            var kernel = new DenseMatrix<float>(size, 1);
 
             float sum = 0;
 
-            float midpoint = (size - 1) / 2f;
+            float midpoint = (size - 1) / 2F;
             for (int i = 0; i < size; i++)
             {
                 float x = i - midpoint;
                 float gx = ImageMaths.Gaussian(x, weight);
                 sum += gx;
-                if (horizontal)
-                {
-                    kernel[0, i] = gx;
-                }
-                else
-                {
-                    kernel[i, 0] = gx;
-                }
+                kernel[0, i] = gx;
             }
 
             // Invert the kernel for sharpening.
             int midpointRounded = (int)midpoint;
-
-            if (horizontal)
+            for (int i = 0; i < size; i++)
             {
-                for (int i = 0; i < size; i++)
+                if (i == midpointRounded)
                 {
-                    if (i == midpointRounded)
-                    {
-                        // Calculate central value
-                        kernel[0, i] = (2F * sum) - kernel[0, i];
-                    }
-                    else
-                    {
-                        // invert value
-                        kernel[0, i] = -kernel[0, i];
-                    }
+                    // Calculate central value
+                    kernel[0, i] = (2F * sum) - kernel[0, i];
                 }
-            }
-            else
-            {
-                for (int i = 0; i < size; i++)
+                else
                 {
-                    if (i == midpointRounded)
-                    {
-                        // Calculate central value
-                        kernel[i, 0] = (2 * sum) - kernel[i, 0];
-                    }
-                    else
-                    {
-                        // invert value
-                        kernel[i, 0] = -kernel[i, 0];
-                    }
+                    // invert value
+                    kernel[0, i] = -kernel[0, i];
                 }
             }
 
             // Normalize kernel so that the sum of all weights equals 1
-            if (horizontal)
-            {
-                for (int i = 0; i < size; i++)
-                {
-                    kernel[0, i] /= sum;
-                }
-            }
-            else
+            for (int i = 0; i < size; i++)
             {
-                for (int i = 0; i < size; i++)
-                {
-                    kernel[i, 0] /= sum;
-                }
+                kernel[0, i] /= sum;
             }
 
             return kernel;
diff --git a/tests/ImageSharp.Tests/Formats/Bmp/BmpDecoderTests.cs b/tests/ImageSharp.Tests/Formats/Bmp/BmpDecoderTests.cs
index 5f2de9f51e..251475567f 100644
--- a/tests/ImageSharp.Tests/Formats/Bmp/BmpDecoderTests.cs
+++ b/tests/ImageSharp.Tests/Formats/Bmp/BmpDecoderTests.cs
@@ -55,6 +55,30 @@ namespace SixLabors.ImageSharp.Tests
             }
         }
 
+        [Theory]
+        [WithFile(WinBmpv2, PixelTypes.Rgba32)]
+        public void BmpDecoder_CanDecodeBmpv2<TPixel>(TestImageProvider<TPixel> provider)
+            where TPixel : struct, IPixel<TPixel>
+        {
+            using (Image<TPixel> image = provider.GetImage(new BmpDecoder()))
+            {
+                image.DebugSave(provider, "png");
+                image.CompareToOriginal(provider);
+            }
+        }
+
+        [Theory]
+        [WithFile(Bit8Palette4, PixelTypes.Rgba32)]
+        public void BmpDecoder_CanDecode4BytePerEntryPalette<TPixel>(TestImageProvider<TPixel> provider)
+            where TPixel : struct, IPixel<TPixel>
+        {
+            using (Image<TPixel> image = provider.GetImage(new BmpDecoder()))
+            {
+                image.DebugSave(provider, "png");
+                image.CompareToOriginal(provider);
+            }
+        }
+
         [Theory]
         [InlineData(Car, 24)]
         [InlineData(F, 24)]
@@ -89,18 +113,17 @@ namespace SixLabors.ImageSharp.Tests
         }
 
         [Theory]
-        [MemberData(nameof(RatioFiles))]
-        public void Identify_VerifyRatio(string imagePath, int xResolution, int yResolution, PixelResolutionUnit resolutionUnit)
+        [WithFile(Os2v2Short, PixelTypes.Rgba32)]
+        public void BmpDecoder_CanDecode_Os2v2XShortHeader<TPixel>(TestImageProvider<TPixel> provider)
+            where TPixel : struct, IPixel<TPixel>
         {
-            var testFile = TestFile.Create(imagePath);
-            using (var stream = new MemoryStream(testFile.Bytes, false))
+            using (Image<TPixel> image = provider.GetImage(new BmpDecoder()))
             {
-                var decoder = new BmpDecoder();
-                IImageInfo image = decoder.Identify(Configuration.Default, stream);
-                ImageMetaData meta = image.MetaData;
-                Assert.Equal(xResolution, meta.HorizontalResolution);
-                Assert.Equal(yResolution, meta.VerticalResolution);
-                Assert.Equal(resolutionUnit, meta.ResolutionUnits);
+                image.DebugSave(provider, "png");
+
+                // TODO: Neither System.Drawing not MagickReferenceDecoder 
+                // can correctly decode this file.
+                // image.CompareToOriginal(provider);
             }
         }
     }
diff --git a/tests/ImageSharp.Tests/Primitives/DenseMatrixTests.cs b/tests/ImageSharp.Tests/Primitives/DenseMatrixTests.cs
index fa4862293f..0af8ae45f9 100644
--- a/tests/ImageSharp.Tests/Primitives/DenseMatrixTests.cs
+++ b/tests/ImageSharp.Tests/Primitives/DenseMatrixTests.cs
@@ -106,5 +106,29 @@ namespace SixLabors.ImageSharp.Tests.Primitives
                 Assert.Equal(0, dense.Data[i]);
             }
         }
+
+        [Fact]
+        public void DenseMatrixCorrectlyCasts()
+        {
+            float[,] actual = new DenseMatrix<float>(FloydSteinbergMatrix);
+            Assert.Equal(FloydSteinbergMatrix, actual);
+        }
+
+        [Fact]
+        public void DenseMatrixCanTranspose()
+        {
+            var dense = new DenseMatrix<int>(3, 1);
+            dense[0, 0] = 1;
+            dense[0, 1] = 2;
+            dense[0, 2] = 3;
+
+            DenseMatrix<int> transposed = dense.Transpose();
+
+            Assert.Equal(dense.Columns, transposed.Rows);
+            Assert.Equal(dense.Rows, transposed.Columns);
+            Assert.Equal(1, transposed[0, 0]);
+            Assert.Equal(2, transposed[1, 0]);
+            Assert.Equal(3, transposed[2, 0]);
+        }
     }
 }
\ No newline at end of file
diff --git a/tests/ImageSharp.Tests/TestImages.cs b/tests/ImageSharp.Tests/TestImages.cs
index 1144a3f7c0..e5b93ab77c 100644
--- a/tests/ImageSharp.Tests/TestImages.cs
+++ b/tests/ImageSharp.Tests/TestImages.cs
@@ -201,6 +201,11 @@ namespace SixLabors.ImageSharp.Tests
             public const string Bit16 = "Bmp/test16.bmp";
             public const string Bit16Inverted = "Bmp/test16-inverted.bmp";
             public const string Bit32Rgb = "Bmp/rgb32.bmp";
+            
+            // Note: This format can be called OS/2 BMPv1, or Windows BMPv2
+            public const string WinBmpv2 = "Bmp/pal8os2v1_winv2.bmp";
+            public const string Bit8Palette4 = "Bmp/pal8-0.bmp";
+            public const string Os2v2Short = "Bmp/pal8os2v2-16.bmp";
 
             public static readonly string[] All
             = {
diff --git a/tests/ImageSharp.Tests/TestUtilities/TestEnvironment.Formats.cs b/tests/ImageSharp.Tests/TestUtilities/TestEnvironment.Formats.cs
index 334b6552ac..7d06847223 100644
--- a/tests/ImageSharp.Tests/TestUtilities/TestEnvironment.Formats.cs
+++ b/tests/ImageSharp.Tests/TestUtilities/TestEnvironment.Formats.cs
@@ -34,8 +34,7 @@ namespace SixLabors.ImageSharp.Tests
         {
             string extension = Path.GetExtension(filePath);
 
-            IImageFormat format = Configuration.ImageFormatsManager.FindFormatByFileExtension(extension);
-            return format;
+            return Configuration.ImageFormatsManager.FindFormatByFileExtension(extension);
         }
 
         private static void ConfigureCodecs(
@@ -69,7 +68,7 @@ namespace SixLabors.ImageSharp.Tests
 
             cfg.ConfigureCodecs(
                 BmpFormat.Instance,
-                SystemDrawingReferenceDecoder.Instance,
+                IsWindows ? (IImageDecoder)SystemDrawingReferenceDecoder.Instance : MagickReferenceDecoder.Instance,
                 bmpEncoder,
                 new BmpImageFormatDetector());
 
diff --git a/tests/ImageSharp.Tests/TestUtilities/Tests/TestEnvironmentTests.cs b/tests/ImageSharp.Tests/TestUtilities/Tests/TestEnvironmentTests.cs
index 30bb16c2a0..122234ae89 100644
--- a/tests/ImageSharp.Tests/TestUtilities/Tests/TestEnvironmentTests.cs
+++ b/tests/ImageSharp.Tests/TestUtilities/Tests/TestEnvironmentTests.cs
@@ -127,7 +127,7 @@ namespace SixLabors.ImageSharp.Tests
 
         [Theory]
         [InlineData("lol/foo.png", typeof(MagickReferenceDecoder))]
-        [InlineData("lol/Rofl.bmp", typeof(SystemDrawingReferenceDecoder))]
+        [InlineData("lol/Rofl.bmp", typeof(MagickReferenceDecoder))]
         [InlineData("lol/Baz.JPG", typeof(JpegDecoder))]
         [InlineData("lol/Baz.gif", typeof(GifDecoder))]
         public void GetReferenceDecoder_ReturnsCorrectDecoders_Linux(string fileName, Type expectedDecoderType)
diff --git a/tests/Images/Input/Bmp/pal8-0.bmp b/tests/Images/Input/Bmp/pal8-0.bmp
new file mode 100644
index 0000000000..ab8815a360
Binary files /dev/null and b/tests/Images/Input/Bmp/pal8-0.bmp differ
diff --git a/tests/Images/Input/Bmp/pal8os2v1_winv2.bmp b/tests/Images/Input/Bmp/pal8os2v1_winv2.bmp
new file mode 100644
index 0000000000..14901b3882
Binary files /dev/null and b/tests/Images/Input/Bmp/pal8os2v1_winv2.bmp differ
diff --git a/tests/Images/Input/Bmp/pal8os2v2-16.bmp b/tests/Images/Input/Bmp/pal8os2v2-16.bmp
new file mode 100644
index 0000000000..95a1d2345a
Binary files /dev/null and b/tests/Images/Input/Bmp/pal8os2v2-16.bmp differ