From 85045d45e2d2e33153ed50f05184d87e436ae8bc Mon Sep 17 00:00:00 2001 From: "wei.lw.li" Date: Tue, 21 Jul 2026 10:05:13 +0800 Subject: [PATCH] =?UTF-8?q?fix(processors):=20=E4=BF=AE=E5=A4=8D=E5=AF=B9?= =?UTF-8?q?=E6=AF=94=E5=BA=A6=E4=B8=8E=E9=98=88=E5=80=BC=E7=AE=97=E5=AD=90?= =?UTF-8?q?16=E4=BD=8D=E5=88=86=E6=94=AF=E7=9A=84=E9=99=8D=E4=BD=8D?= =?UTF-8?q?=E7=B2=BE=E5=BA=A6=E6=8D=9F=E5=A4=B1?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - ContrastProcessor: CLAHE分支改用原生CvInvoke.CLAHE(支持CV_16U)，不再16→8→16往返 - ThresholdProcessor: Otsu分支在16位完整直方图(65536 bin)上原生计算最优阈值 --- .../图像变换/ThresholdProcessor.cs | 77 ++++++++++-- .../图像增强/ContrastProcessor.cs | 115 +----------------- 2 files changed, 70 insertions(+), 122 deletions(-) diff --git a/XP.ImageProcessing.Processors/图像变换/ThresholdProcessor.cs b/XP.ImageProcessing.Processors/图像变换/ThresholdProcessor.cs index 6c1c7242..74d8d250 100644 --- a/XP.ImageProcessing.Processors/图像变换/ThresholdProcessor.cs +++ b/XP.ImageProcessing.Processors/图像变换/ThresholdProcessor.cs @@ -71,25 +71,29 @@ public class ThresholdProcessor : ImageProcessorBase if (useOtsu) { - // Otsu 仅支持 8 位，16 位需先降位 - if (typeof(TDepth) == typeof(ushort)) - { - using var img8 = PixelDepthHelper.ToByteImage(inputImage); - using var res8 = new Image(inputImage.Size); - CvInvoke.Threshold(img8, res8, minThreshold >> 8, 255, ThresholdType.Otsu); - // 将 8 位二值结果升位到 16 位 - var res16 = PixelDepthHelper.FromByteImage(res8); - _logger.Debug("Process: UseOtsu=true (16→8→16)"); - return res16; - } - else + // 8 位直接用 OpenCV Otsu；16 位在完整位深直方图上原生计算 Otsu 阈值，避免降位 + if (typeof(TDepth) == typeof(byte)) { using var res8 = new Image(inputImage.Size); CvInvoke.Threshold(inputImage as Image ?? inputImage.Convert(), res8, minThreshold, 255, ThresholdType.Otsu); - _logger.Debug("Process: UseOtsu=true"); + _logger.Debug("Process: UseOtsu=true (8bit native)"); return res8 as Image ?? PixelDepthHelper.FromByteImage(res8); } + else + { + int otsuThreshold = ComputeOtsuThreshold16((inputImage as Image)!, maxVal); + Parallel.For(0, height, y => + { + for (int x = 0; x < width; x++) + { + int val = PixelDepthHelper.ReadPixel(inputImage, y, x); + PixelDepthHelper.WritePixel(result, y, x, val > otsuThreshold ? maxVal : 0); + } + }); + _logger.Debug("Process: UseOtsu=true (16bit native), threshold={Threshold}", otsuThreshold); + return result; + } } // 手工双阈值分割（支持全位深） @@ -106,4 +110,51 @@ public class ThresholdProcessor : ImageProcessorBase _logger.Debug("Process: MinThreshold={Min}, MaxThreshold={Max}", minThreshold, maxThreshold); return result; } + + /// + /// 在完整 16 位直方图（65536 bin）上计算 Otsu 最优阈值（最大化类间方差）。 + /// 返回值为灰度阈值：像素值 > 阈值 判为前景。 + /// + private static int ComputeOtsuThreshold16(Image image, int maxVal) + { + int levels = maxVal + 1; + var histogram = new long[levels]; + int h = image.Height, w = image.Width; + var data = image.Data; + for (int y = 0; y < h; y++) + for (int x = 0; x < w; x++) + histogram[data[y, x, 0]]++; + + long totalPixels = (long)w * h; + double totalSum = 0; + for (int i = 0; i < levels; i++) + totalSum += (double)i * histogram[i]; + + double bgSum = 0; + long bgPixels = 0; + double maxVariance = -1; + int bestThreshold = 0; + + for (int t = 0; t < levels; t++) + { + bgPixels += histogram[t]; + if (bgPixels == 0) continue; + + long fgPixels = totalPixels - bgPixels; + if (fgPixels == 0) break; + + bgSum += (double)t * histogram[t]; + double bgMean = bgSum / bgPixels; + double fgMean = (totalSum - bgSum) / fgPixels; + double variance = (double)bgPixels * fgPixels * (bgMean - fgMean) * (bgMean - fgMean); + + if (variance > maxVariance) + { + maxVariance = variance; + bestThreshold = t; + } + } + + return bestThreshold; + } } \ No newline at end of file diff --git a/XP.ImageProcessing.Processors/图像增强/ContrastProcessor.cs b/XP.ImageProcessing.Processors/图像增强/ContrastProcessor.cs index 5bab58d0..fa72094b 100644 --- a/XP.ImageProcessing.Processors/图像增强/ContrastProcessor.cs +++ b/XP.ImageProcessing.Processors/图像增强/ContrastProcessor.cs @@ -95,19 +95,12 @@ public class ContrastProcessor : ImageProcessorBase if (useCLAHE) { - // CLAHE 仅对 8 位有原生 CvInvoke 支持，16 位降位处理 - if (typeof(TDepth) == typeof(ushort)) - { - using var img8 = PixelDepthHelper.ToByteImage(inputImage); - using var res8 = ApplyCLAHE(img8, clipLimit); - _logger.Debug("Process (CLAHE 16→8→16)"); - return PixelDepthHelper.FromByteImage(res8); - } - else - { - var res = ApplyCLAHE(inputImage as Image ?? inputImage.Convert(), clipLimit); - return res as Image ?? PixelDepthHelper.FromByteImage(res); - } + // OpenCV CLAHE 原生支持 CV_8UC1 和 CV_16UC1，直接按位深处理，无需降位 + var result = new Image(inputImage.Size); + CvInvoke.CLAHE(inputImage, clipLimit, new Size(8, 8), result); + _logger.Debug("Process (CLAHE native {Depth}bit): ClipLimit={ClipLimit}", + typeof(TDepth) == typeof(ushort) ? 16 : 8, clipLimit); + return result; } else if (autoContrast) { @@ -140,100 +133,4 @@ public class ContrastProcessor : ImageProcessorBase _logger.Debug("AutoContrastStretch: min={Min}, max={Max}", minVal, maxVal); return PixelDepthHelper.FromFloatImage(floatImage); } - - private Image ApplyCLAHE(Image inputImage, double clipLimit) - { - int tileSize = 8; - int width = inputImage.Width; - int height = inputImage.Height; - byte[,,] srcData = inputImage.Data; - - // 计算分块数 - int tilesX = (width + tileSize - 1) / tileSize; - int tilesY = (height + tileSize - 1) / tileSize; - int actualTileW = (width + tilesX - 1) / tilesX; - int actualTileH = (height + tilesY - 1) / tilesY; - - // 为每个 tile 计算带 clip limit 的均衡化映射表 - var luts = new byte[tilesY, tilesX, 256]; - for (int ty = 0; ty < tilesY; ty++) - { - for (int tx = 0; tx < tilesX; tx++) - { - int x0 = tx * actualTileW; - int y0 = ty * actualTileH; - int x1 = Math.Min(x0 + actualTileW, width); - int y1 = Math.Min(y0 + actualTileH, height); - int tilePixels = (x1 - x0) * (y1 - y0); - - // 构建直方图 - var hist = new int[256]; - for (int y = y0; y < y1; y++) - for (int x = x0; x < x1; x++) - hist[srcData[y, x, 0]]++; - - // Clip limit 裁剪并重新分配 - int clipThreshold = (int)(clipLimit * tilePixels / 256); - if (clipThreshold > 0) - { - int excess = 0; - for (int i = 0; i < 256; i++) - { - if (hist[i] > clipThreshold) - { - excess += hist[i] - clipThreshold; - hist[i] = clipThreshold; - } - } - int avgInc = excess / 256; - int remainder = excess - avgInc * 256; - for (int i = 0; i < 256; i++) - hist[i] += avgInc + (i < remainder ? 1 : 0); - } - - // 构建 CDF 映射表 - int sum = 0; - for (int i = 0; i < 256; i++) - { - sum += hist[i]; - luts[ty, tx, i] = (byte)Math.Clamp(sum * 255 / tilePixels, 0, 255); - } - } - } - - // 双线性插值生成结果 - var result = new Image(width, height); - byte[,,] dstData = result.Data; - - for (int y = 0; y < height; y++) - { - for (int x = 0; x < width; x++) - { - // 计算当前像素所在 tile 的中心坐标 - double fx = (double)x / actualTileW - 0.5; - double fy = (double)y / actualTileH - 0.5; - int tx0 = Math.Max(0, (int)fx); - int ty0 = Math.Max(0, (int)fy); - int tx1 = Math.Min(tx0 + 1, tilesX - 1); - int ty1 = Math.Min(ty0 + 1, tilesY - 1); - double ax = fx - tx0; - double ay = fy - ty0; - ax = Math.Clamp(ax, 0, 1); - ay = Math.Clamp(ay, 0, 1); - - byte val = srcData[y, x, 0]; - double v00 = luts[ty0, tx0, val]; - double v10 = luts[ty0, tx1, val]; - double v01 = luts[ty1, tx0, val]; - double v11 = luts[ty1, tx1, val]; - - double interpolated = v00 * (1 - ax) * (1 - ay) + v10 * ax * (1 - ay) - + v01 * (1 - ax) * ay + v11 * ax * ay; - dstData[y, x, 0] = (byte)Math.Clamp((int)(interpolated + 0.5), 0, 255); - } - } - - _logger.Debug("ApplyCLAHE: ClipLimit={ClipLimit}, TileSize={TileSize}", clipLimit, tileSize); - return result; - } }