当前位置: 首页 > news >正文

图像保边滤波之BEEPS滤波算法

目录

1 简介

2 算法原理

3 代码实现

4 演示Demo

4.1 开发环境

4.2 功能介绍

4.3 下载地址

参考


1 简介

        BEEPS(Bias Elimination in Edge-Preserving Smoothing) 是一种基于偏微分方程(PDE)的边缘保留平滑滤波算法。它能够在平滑图像的同时有效消除偏差(Bias),从而更好地保留边缘和细节。BEEPS 滤波算法广泛应用于图像去噪、图像增强和医学图像处理等领域。

2 算法原理

        BEEPS 滤波的核心思想是通过引入偏差消除项,改进传统的边缘保留平滑算法(如各向异性扩散)。其数学模型基于以下偏微分方程:

\frac{\vartheta{I}}{\vartheta{t}} = div(c(|\triangledown I|)* \triangledown I) - \lambda (I-I_0)

        其中

  • I 是图像。

  • I 是图像的梯度。

  • c(∣∇I∣) 是扩散系数,控制滤波的强度。

  • \lambda是偏差消除项的权重。

  • I_0是原始图像

        扩散系数c(∣∇I∣) 通常定义为:

c(|\triangledown I|) = \frac{1}{1+(\frac{|\triangledown I|}{k})^2}

        或

c(|\triangledown I|) = -exp(-(\frac{|\triangledown I|}{k})^2)

        其中,k 是一个控制边缘敏感度的参数。

        算法步骤:

  1. 计算图像的梯度 ∇I

  2. 根据梯度计算扩散系数 c(∣∇I∣)。

  3. 使用扩散系数更新图像像素值。

  4. 重复上述步骤,直到达到指定的迭代次数。

3 代码实现

        Anisotropic滤波算法的的C语言实现代码如下

double Gaussian(int u, int v, double sigma)
{int t = -(u - v) * (u - v);return exp((double)t / sigma);
}
int BEEPSHorizontal(unsigned char* srcPtr, int width, int height, unsigned char* outData, double sigma, int c)
{unsigned char* F = (unsigned char*)malloc(sizeof(unsigned char) * width * height);int *s = (int*)malloc(sizeof(int) * width);int *v = (int*)malloc(sizeof(int) * width);int pos = 0, X = 0, Y = 0;int p = 0;memset(F, 0, width * height);memset(outData, 0, width * height);memset(s, 0, width);memset(v, 0, width);unsigned char* D = outData;for (int y = 0; y < height; y++){for (int x = 0; x < width; x++){X = width - 1 - x;Y = height - 1 - y;if (x == 0){pos = x + y * width;F[pos] = srcPtr[pos];s[0] = srcPtr[pos];p = X;pos = p + Y * width;v[p] = srcPtr[pos];D[pos] = srcPtr[pos];}else{p = x;pos = p + y * width;s[p] = (int)(10.0 * Gaussian(srcPtr[pos], F[pos - 1], sigma));F[pos] = CLIP3((((100 - s[p] * c) * srcPtr[pos] + s[p] * c * F[pos - 1]) / 100), 0, 255);p = X;pos = p + Y * width;v[p] = (int)(10.0 * Gaussian(srcPtr[pos], D[pos + 1], sigma));D[pos] = CLIP3((((100 - v[p] * c) * srcPtr[pos] + v[p] * c * D[pos + 1]) / 100), 0, 255);}}}for (int i = 0; i < height * width; i++){D[i] = CLIP3(((10 * F[i] - (10 - c) * (srcPtr[i]) + 10 * D[i]) / (10 + c)), 0, 255);}free(F);free(s);free(v);return 0;
}int BEEPSVertical(unsigned char* srcPtr, int width, int height, unsigned char* outData, double sigma, int c)
{unsigned char* F = (unsigned char*)malloc(sizeof(unsigned char) * width * height);unsigned char* D = outData;int* s = (int*)malloc(sizeof(int) * height);int* v = (int*)malloc(sizeof(int) * height);int pos = 0, X = 0, Y = 0;memset(s, 0, height);memset(v, 0, height);for (int x = 0; x < width; x++){for (int y = 0; y < height; y++){X = width - 1 - x;Y = height - 1 - y;if (y == 0){pos = x + y * width;F[pos] = srcPtr[pos];s[y] = srcPtr[pos];pos = X + Y * width;D[pos] = srcPtr[pos];v[Y] = srcPtr[pos];}else{pos = x + y * width;s[y] = (int)(10.0 * Gaussian(srcPtr[pos], F[pos - width], sigma));F[pos] = CLIP3((((100 - s[y] * c) * srcPtr[pos] + s[y] * c * F[pos - width]) / 100), 0, 255);pos = X + Y * width;v[Y] = (int)(10.0 * Gaussian(srcPtr[pos], D[pos + width], sigma));D[pos] = CLIP3((((100 - v[Y] * c) * srcPtr[pos] + v[Y] * c * D[pos + width]) / 100), 0, 255);}}}for (int i = 0; i < height*width; i++){D[i] = CLIP3(((10 * F[i] - (10 - c) * (srcPtr[i]) + 10 * D[i]) / (10 + c)), 0, 255);}free(F);free(s);free(v);return 0;
}void BEEPSProcess(unsigned char* srcPtr, int width, int height, float sigma, float c)
{float* GMAP = (float*)malloc(sizeof(float) * 256 * 256);for (int j = 0; j < 256; j++){for (int i = 0; i < 256; i++){GMAP[i + j * 256] = Gaussian(i, j, sigma);}}sigma = sigma > 50 ? 50 : sigma;sigma = sigma * sigma * 2.0f;float Lamba = 10.0f * (float)(1 - (sqrt(2.0f * c * c + 1) - 1) / (c * c));unsigned char* pSrc = srcPtr;unsigned char* hValue = (unsigned char*)malloc(sizeof(unsigned char) * width * height);unsigned char* vValue = (unsigned char*)malloc(sizeof(unsigned char) * width * height);unsigned char* dstValue = (unsigned char*)malloc(sizeof(unsigned char) * width * height);BEEPSHorizontal(pSrc, width, height, hValue, sigma, Lamba);BEEPSVertical(hValue, width, height, vValue, sigma, Lamba);BEEPSVertical(pSrc, width, height, hValue, sigma, Lamba);BEEPSHorizontal(hValue, width, height, dstValue, sigma, Lamba);for (int i = 0; i < width * height; i++){*pSrc++ = CLIP3(((vValue[i] + dstValue[i]) / 2), 0, 255);}free(hValue);free(vValue);free(dstValue);
}int mxBeepsFilter(unsigned char* srcData, int nWidth, int nHeight, int nStride, float delta, float delta_s)
{if (srcData == NULL){return 0;}if (delta == 0 || delta_s == 0)return 0;unsigned char* yData = (unsigned char*)malloc(sizeof(unsigned char) * nWidth * nHeight);unsigned char* cbData = (unsigned char*)malloc(sizeof(unsigned char) * nWidth * nHeight);unsigned char* crData = (unsigned char*)malloc(sizeof(unsigned char) * nWidth * nHeight);unsigned char* pSrc = srcData;int Y, CB, CR;unsigned char* pY = yData;unsigned char* pCb = cbData;unsigned char* pCr = crData;for (int j = 0; j < nHeight; j++){for (int i = 0; i < nWidth; i++){RGBToYCbCr(pSrc[2], pSrc[1], pSrc[0], &Y, &CB, &CR);*pY = Y;*pCb = CB;*pCr = CR;pY++;pCb++;pCr++;pSrc += 4;}}BEEPSProcess(yData, nWidth, nHeight, delta, delta_s);pSrc = srcData;pY = yData;pCb = cbData;pCr = crData;int R, G, B;for (int j = 0; j < nHeight; j++){for (int i = 0; i < nWidth; i++){YCbCrToRGB(*pY, *pCb, *pCr, &R, &G, &B);pSrc[0] = B;pSrc[1] = G;pSrc[2] = R;pY++;pCb++;pCr++;pSrc += 4;}}free(yData);free(cbData);free(crData);return 0;
}

4 演示Demo

4.1 开发环境

  • Windows 10 Pro x64

  • Visual Studio 2015

4.2 功能介绍

        演示程序主界面如下图所示,具有图像读取、显示、保存、双边滤波、表面模糊、导向滤波、局部均值方差滤波、各向异性扩散滤波、Smart Blur滤波、MeanShift滤波、BEEPS滤波、处理耗时等功能

原图

滤波效果图

4.3 下载地址

        开发环境:

  • Windows 10 pro x64

  • Visual Studio 2015

        下载地址:图像保边滤波之BEEPS滤波算法Demo

参考

        图像视频滤镜与人像美颜美妆算法详解. 胡耀武、谭娟、李云夕. 电子工业出版社、2020-07

http://www.xdnf.cn/news/188785.html

相关文章:

  • KUKA机器人自动备份设置
  • vscode 使用gitcode团队管理项目
  • 区块链随学随记
  • jetson nano上Ubuntu系统调用摄像头bug
  • 塔能科技:点亮节能之光,赋能工厂与城市
  • 20250428-AI Agent:智能体的演进与未来
  • 包装产线通过canopen转Profinet网关控制伺服
  • 关于常量指针和指向常量的指针
  • 泰山派常用命令
  • map和set:
  • ai环境conda带torch整体迁移。
  • 一文了解 模型上下文协议(MCP)
  • word插入APA格式的参考文献
  • NGINX ngx_http_addition_module 模块响应体前后注入内容
  • VS2022+OpenCasCade配置编译
  • 【leetcode】最长公共子路径问题
  • 从大众传媒到数字生态:开源AI智能名片链动2+1模式S2B2C商城小程序驱动的营销革命
  • prompt提示词编写技巧
  • Context7 MCP:提供实时、版本特定的文档以解决AI幻觉问题
  • Go 语言入门:(一) 环境安装
  • 大语言模型(LLMs)微调技术总结
  • web 基础与 http 协议
  • 【JAVA ee初阶】多线程(3)
  • es+kibana---集群部署
  • MYOJ_1349:(洛谷P3951)[NOIP 2017 提高组] 小凯的疑惑(数学公式套用,两步搞定代码)
  • 快速上手QEMU:创建你的第一个虚拟机实例
  • 深入浅出限流算法(一):简单但有“坑”的固定窗口计数器
  • 大数据应用开发和项目实战
  • 第五章、SpringBoot与消息通信(三)
  • 线性代数——行列式⭐