From 2dcfcda850a930c25d606675a4e7e58a34c89b41 Mon Sep 17 00:00:00 2001
From: sinvo <admin@mcraft.cn>
Date: Thu, 19 Mar 2026 16:34:37 +0800
Subject: [PATCH] =?UTF-8?q?=E4=BF=AE=E5=A4=8D=E6=96=87=E6=A1=A3=E8=BE=B9?=
 =?UTF-8?q?=E7=BC=98=E9=98=B4=E5=BD=B1=E5=8F=98=E7=BA=AF=E9=BB=91=E7=9A=84?=
 =?UTF-8?q?=E9=97=AE=E9=A2=98?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

通过背景亮度中位数计算阴影阈值，背景估计值低于中位数60%的区域
识别为拍照阴影，直接输出白色而非当作墨迹处理成黑色。

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 CamScanner.cs | 35 +++++++++++++++++++++++++++++++++++
 1 file changed, 35 insertions(+)

diff --git a/CamScanner.cs b/CamScanner.cs
index 271df07..00b2d70 100644
--- a/CamScanner.cs
+++ b/CamScanner.cs
@@ -376,8 +376,43 @@ public static class DocumentScanner
 
         // 增益：让95百分位的墨迹深度映射到接近纯黑
         // 用非线性映射（幂函数）让浅墨迹也能变深
+        //
+        // 阴影识别：拍照时文档边缘常有阴影，特征是：
+        //   - 原始灰度很低（整体偏暗）
+        //   - 背景估计也偏低（因为阴影区域大面积偏暗）
+        //   - 但 bg - gray 的差值也可能很大
+        // 区分方法：如果背景估计值本身就很低（< 阈值），说明是阴影区域，
+        // 应该输出白色而非黑色
+        //
+        // 同时计算背景亮度的中位数，作为"正常纸面亮度"参考
+        int[] bgHist = new int[256];
+        for (int i = 0; i < bgData.Length; i++)
+        {
+            bgHist[bgData[i]]++;
+        }
+        int bgCum = 0;
+        int bgMedian = 128;
+        for (int i = 0; i < 256; i++)
+        {
+            bgCum += bgHist[i];
+            if (bgCum >= totalPixels / 2)
+            {
+                bgMedian = i;
+                break;
+            }
+        }
+        // 阴影阈值：背景亮度低于中位数的60%视为阴影区域
+        int shadowThresh = (int)(bgMedian * 0.6);
+
         for (int i = 0; i < grayData.Length; i++)
         {
+            // 阴影检测：背景估计值很低 → 阴影区域 → 输出白色
+            if (bgData[i] < shadowThresh)
+            {
+                resultData[i] = 255;
+                continue;
+            }
+
             int ink = bgData[i] - grayData[i];
             if (ink < 0) ink = 0;