半径无关单核单线程最快速高斯模糊实现(附完整C代码)
之前,俺也发过不少快速高斯模糊算法.
俺一般认为,只要处理一千六百万像素彩色图片,在2.2GHz的CPU上单核单线程超过1秒的算法,都是不快的.
之前发的几个算法,在俺2.2GHz的CPU上耗时都会超过1秒.
而众所周知,快速高斯模糊有很多实现方法:
1.FIR (Finite impulse response)
https://zh.wikipedia.org/wiki/%E9%AB%98%E6%96%AF%E6%A8%A1%E7%B3%8A
2.SII (Stacked integral images)
http://dx.doi.org/10.1109/ROBOT.2010.5509400
http://arxiv.org/abs/1107.4958
3.Vliet-Young-Verbeek (Recursive filter)
http://dx.doi.org/10.1016/0165-1684(95)00020-E
http://dx.doi.org/10.1109/ICPR.1998.711192
4.DCT (Discrete Cosine Transform)
http://dx.doi.org/10.1109/78.295213
5.box (Box filter)
http://dx.doi.org/10.1109/TPAMI.1986.4767776
6.AM(Alvarez, Mazorra)
http://www.jstor.org/stable/2158018
7.Deriche (Recursive filter)
http://hal.inria.fr/docs/00/07/47/78/PDF/RR-1893.pdf
8.ebox (Extended Box)
http://dx.doi.org/10.1007/978-3-642-24785-9_38
9.IIR (Infinite Impulse Response)
https://software.intel.com/zh-cn/articles/iir-gaussian-blur-filter-implementation-using-intel-advanced-vector-extensions
10.FA (Fast Anisotropic)
http://mathinfo.univ-reims.fr/IMG/pdf/Fast_Anisotropic_Gquss_Filtering_-_GeusebroekECCV02.pdf
......
实现高斯模糊的方法虽然很多,但是作为算法而言,核心关键是简单高效.
目前俺经过实测,IIR是兼顾效果以及性能的不错的方法,也是半径无关(即模糊不同强度耗时基本不变)的实现.
英特尔官方实现的这份:
IIR Gaussian Blur Filter Implementation using Intel® Advanced Vector Extensions [PDF 513KB]
source: gaussian_blur.cpp [36KB]
采用了英特尔处理器的流(SIMD)指令,算法处理速度极其惊人.
俺写算法追求干净整洁,高效简单,换言之就是不采用任何硬件加速方案,实现简单高效,以适应不同硬件环境.
故基于英特尔这份代码,俺对其进行了改写以及优化.
最终在俺2.20GHz的CPU上,单核单线程,不采用流(SIMD)指令,达到了,处理一千六百万像素的彩色照片仅需700毫秒左右.
按照惯例,还是贴个效果图比较直观.
之前也有网友问过这个算法的实现问题.
想了想,还是将代码共享出来,供大家参考学习.
完整代码:
void CalGaussianCoeff(float sigma, float * a0, float * a1, float * a2, float * a3, float * b1, float * b2, float * cprev, float * cnext) {
float alpha, lamma, k;
if (sigma < 0.5f)
sigma = 0.5f;
alpha = (float)exp((0.726) * (0.726)) / sigma;
lamma = (float)exp(-alpha);
*b2 = (float)exp(-2 * alpha);
k = (1 - lamma) * (1 - lamma) / (1 + 2 * alpha * lamma - (*b2));
*a0 = k; *a1 = k * (alpha - 1) * lamma;
*a2 = k * (alpha + 1) * lamma;
*a3 = -k * (*b2);
*b1 = -2 * lamma;
*cprev = (*a0 + *a1) / (1 + *b1 + *b2);
*cnext = (*a2 + *a3) / (1 + *b1 + *b2);
}
void gaussianHorizontal(unsigned char * bufferPerLine, unsigned char * pRowInitial, unsigned char * pColumn, int Width, int Height, int Channels, int Nwidth, int a0a1, int a2a3, int b1b2, int cprev, int cnext)
{
int HeightStep = Channels*Height;
int lastWidth = Width - 1;
if (Channels == 3)
{
int prevOut[3];
prevOut[0] = (pRowInitial[0] * cprev) >> 8;
prevOut[1] = (pRowInitial[1] * cprev) >> 8;
prevOut[2] = (pRowInitial[2] * cprev) >> 8;
for (int x = 0; x < Width; ++x) {
prevOut[0] = ((pRowInitial[0] * (a0a1)) - (prevOut[0] * (b1b2))) >> 16;
prevOut[1] = ((pRowInitial[1] * (a0a1)) - (prevOut[1] * (b1b2))) >> 16;
prevOut[2] = ((pRowInitial[2] * (a0a1)) - (prevOut[2] * (b1b2))) >> 16;
bufferPerLine[0] = prevOut[0];
bufferPerLine[1] = prevOut[1];
bufferPerLine[2] = prevOut[2];
bufferPerLine += Channels;
pRowInitial += Channels;
}
pRowInitial -= Channels;
pColumn += HeightStep * lastWidth;
bufferPerLine -= Channels;
prevOut[0] = (pRowInitial[0] * cnext) >> 8;
prevOut[1] = (pRowInitial[1] * cnext) >> 8;
prevOut[2] = (pRowInitial[2] * cnext) >> 8;
for (int x = lastWidth; x >= 0; --x) {
prevOut[0] = ((pRowInitial[0] * (a2a3)) - (prevOut[0] * (b1b2))) >> 16;
prevOut[1] = ((pRowInitial[1] * (a2a3)) - (prevOut[1] * (b1b2))) >> 16;
prevOut[2] = ((pRowInitial[2] * (a2a3)) - (prevOut[2] * (b1b2))) >> 16;
bufferPerLine[0] += prevOut[0];
bufferPerLine[1] += prevOut[1];
bufferPerLine[2] += prevOut[2];
pColumn[0] = bufferPerLine[0];
pColumn[1] = bufferPerLine[1];
pColumn[2] = bufferPerLine[2];
pRowInitial -= Channels;
pColumn -= HeightStep;
bufferPerLine -= Channels;
}
}
else if (Channels == 4)
{
int prevOut[4];
prevOut[0] = (pRowInitial[0] * cprev) >> 8;
prevOut[1] = (pRowInitial[1] * cprev) >> 8;
prevOut[2] = (pRowInitial[2] * cprev) >> 8;
prevOut[3] = (pRowInitial[3] * cprev) >> 8;
for (int x = 0; x < Width; ++x) {
prevOut[0] = ((pRowInitial[0] * (a0a1)) - (prevOut[0] * (b1b2))) >> 16;
prevOut[1] = ((pRowInitial[1] * (a0a1)) - (prevOut[1] * (b1b2))) >> 16;
prevOut[2] = ((pRowInitial[2] * (a0a1)) - (prevOut[2] * (b1b2))) >> 16;
prevOut[3] = ((pRowInitial[3] * (a0a1)) - (prevOut[3] * (b1b2))) >> 16;
bufferPerLine[0] = prevOut[0];
bufferPerLine[1] = prevOut[1];
bufferPerLine[2] = prevOut[2];
bufferPerLine[3] = prevOut[3];
bufferPerLine += Channels;
pRowInitial += Channels;
}
pRowInitial -= Channels;
pColumn += HeightStep * lastWidth;
bufferPerLine -= Channels;
prevOut[0] = (pRowInitial[0] * cnext) >> 8;
prevOut[1] = (pRowInitial[1] * cnext) >> 8;
prevOut[2] = (pRowInitial[2] * cnext) >> 8;
prevOut[3] = (pRowInitial[3] * cnext) >> 8;
for (int x = lastWidth; x >= 0; --x) {
prevOut[0] = ((pRowInitial[0] * a2a3) - (prevOut[0] * b1b2)) >> 16;
prevOut[1] = ((pRowInitial[1] * a2a3) - (prevOut[1] * b1b2)) >> 16;
prevOut[2] = ((pRowInitial[2] * a2a3) - (prevOut[2] * b1b2)) >> 16;
prevOut[3] = ((pRowInitial[3] * a2a3) - (prevOut[3] * b1b2)) >> 16;
bufferPerLine[0] += prevOut[0];
bufferPerLine[1] += prevOut[1];
bufferPerLine[2] += prevOut[2];
bufferPerLine[3] += prevOut[3];
pColumn[0] = bufferPerLine[0];
pColumn[1] = bufferPerLine[1];
pColumn[2] = bufferPerLine[2];
pColumn[3] = bufferPerLine[3];
pRowInitial -= Channels;
pColumn -= HeightStep;
bufferPerLine -= Channels;
}
}
else if (Channels == 1)
{
int prevOut = (pRowInitial[0] * cprev) >> 8;
for (int x = 0; x < Width; ++x) {
prevOut = ((pRowInitial[0] * (a0a1)) - (prevOut * (b1b2))) >> 16;
bufferPerLine[0] = prevOut;
bufferPerLine += Channels;
pRowInitial += Channels;
}
pRowInitial -= Channels;
pColumn += HeightStep*lastWidth;
bufferPerLine -= Channels;
prevOut = (pRowInitial[0] * cnext) >> 8;
for (int x = lastWidth; x >= 0; --x) {
prevOut = ((pRowInitial[0] * a2a3) - (prevOut * b1b2)) >> 16;;
bufferPerLine[0] += prevOut;
pColumn[0] = bufferPerLine[0];
pRowInitial -= Channels;
pColumn -= HeightStep;
bufferPerLine -= Channels;
}
}
}
void gaussianVertical(unsigned char * bufferPerLine, unsigned char * pRowInitial, unsigned char * pColInitial, int Height, int Width, int Channels, int a0a1, int a2a3, int b1b2, int cprev, int cnext) {
int WidthStep = Channels*Width;
int lastHeight = Height - 1;
if (Channels == 3)
{
int prevOut[3];
prevOut[0] = (pRowInitial[0] * cprev) >> 8;
prevOut[1] = (pRowInitial[1] * cprev) >> 8;
prevOut[2] = (pRowInitial[2] * cprev) >> 8;
for (int y = 0; y < Height; y++) {
prevOut[0] = ((pRowInitial[0] * a0a1) - (prevOut[0] * b1b2)) >> 16;
prevOut[1] = ((pRowInitial[1] * a0a1) - (prevOut[1] * b1b2)) >> 16;
prevOut[2] = ((pRowInitial[2] * a0a1) - (prevOut[2] * b1b2)) >> 16;
bufferPerLine[0] = prevOut[0];
bufferPerLine[1] = prevOut[1];
bufferPerLine[2] = prevOut[2];
bufferPerLine += Channels;
pRowInitial += Channels;
}
pRowInitial -= Channels;
bufferPerLine -= Channels;
pColInitial += WidthStep * lastHeight;
prevOut[0] = (pRowInitial[0] * cnext) >> 8;
prevOut[1] = (pRowInitial[1] * cnext) >> 8;
prevOut[2] = (pRowInitial[2] * cnext) >> 8;
for (int y = lastHeight; y >= 0; y--) {
prevOut[0] = ((pRowInitial[0] * a2a3) - (prevOut[0] * b1b2)) >> 16;
prevOut[1] = ((pRowInitial[1] * a2a3) - (prevOut[1] * b1b2)) >> 16;
prevOut[2] = ((pRowInitial[2] * a2a3) - (prevOut[2] * b1b2)) >> 16;
bufferPerLine[0] += prevOut[0];
bufferPerLine[1] += prevOut[1];
bufferPerLine[2] += prevOut[2];
pColInitial[0] = bufferPerLine[0];
pColInitial[1] = bufferPerLine[1];
pColInitial[2] = bufferPerLine[2];
pRowInitial -= Channels;
pColInitial -= WidthStep;
bufferPerLine -= Channels;
}
}
else if (Channels == 4)
{
int prevOut[4];
prevOut[0] = (pRowInitial[0] * cprev) >> 8;
prevOut[1] = (pRowInitial[1] * cprev) >> 8;
prevOut[2] = (pRowInitial[2] * cprev) >> 8;
prevOut[3] = (pRowInitial[3] * cprev) >> 8;
for (int y = 0; y < Height; y++) {
prevOut[0] = ((pRowInitial[0] * a0a1) - (prevOut[0] * b1b2)) >> 16;
prevOut[1] = ((pRowInitial[1] * a0a1) - (prevOut[1] * b1b2)) >> 16;
prevOut[2] = ((pRowInitial[2] * a0a1) - (prevOut[2] * b1b2)) >> 16;
prevOut[3] = ((pRowInitial[3] * a0a1) - (prevOut[3] * b1b2)) >> 16;
bufferPerLine[0] = prevOut[0];
bufferPerLine[1] = prevOut[1];
bufferPerLine[2] = prevOut[2];
bufferPerLine[3] = prevOut[3];
bufferPerLine += Channels;
pRowInitial += Channels;
}
pRowInitial -= Channels;
bufferPerLine -= Channels;
pColInitial += WidthStep*lastHeight;
prevOut[0] = (pRowInitial[0] * cnext) >> 8;
prevOut[1] = (pRowInitial[1] * cnext) >> 8;
prevOut[2] = (pRowInitial[2] * cnext) >> 8;
prevOut[3] = (pRowInitial[3] * cnext) >> 8;
for (int y = lastHeight; y >= 0; y--) {
prevOut[0] = ((pRowInitial[0] * a2a3) - (prevOut[0] * b1b2)) >> 16;
prevOut[1] = ((pRowInitial[1] * a2a3) - (prevOut[1] * b1b2)) >> 16;
prevOut[2] = ((pRowInitial[2] * a2a3) - (prevOut[2] * b1b2)) >> 16;
prevOut[3] = ((pRowInitial[3] * a2a3) - (prevOut[3] * b1b2)) >> 16;
bufferPerLine[0] += prevOut[0];
bufferPerLine[1] += prevOut[1];
bufferPerLine[2] += prevOut[2];
bufferPerLine[3] += prevOut[3];
pColInitial[0] = bufferPerLine[0];
pColInitial[1] = bufferPerLine[1];
pColInitial[2] = bufferPerLine[2];
pColInitial[3] = bufferPerLine[3];
pRowInitial -= Channels;
pColInitial -= WidthStep;
bufferPerLine -= Channels;
}
}
else if (Channels == 1)
{
int prevOut = 0;
prevOut = (pRowInitial[0] * cprev) >> 8;
for (int y = 0; y < Height; y++) {
prevOut = ((pRowInitial[0] * a0a1) - (prevOut * b1b2)) >> 16;
bufferPerLine[0] = prevOut;
bufferPerLine += Channels;
pRowInitial += Channels;
}
pRowInitial -= Channels;
bufferPerLine -= Channels;
pColInitial += WidthStep*lastHeight;
prevOut = (pRowInitial[0] * cnext) >> 8;
for (int y = lastHeight; y >= 0; y--) {
prevOut = ((pRowInitial[0] * a2a3) - (prevOut * b1b2)) >> 16;
bufferPerLine[0] += prevOut;
pColInitial[0] = bufferPerLine[0];
pRowInitial -= Channels;
pColInitial -= WidthStep;
bufferPerLine -= Channels;
}
}
}
//本人博客:http://tntmonks.cnblogs.com/ 转载请注明出处.
void GaussianBlurFilter(unsigned char * inputBuffer, unsigned char * outputBuffer, int Width, int Height, int Channels, float gaussianSigma = 2.0f) {
float a0, a1, a2, a3, b1, b2, cprev, cnext;
CalGaussianCoeff(gaussianSigma, &a0, &a1, &a2, &a3, &b1, &b2, &cprev, &cnext);
int icprev = cprev * 256;
int icnext = cnext * 256;
int a0a1 = (a0 + a1) * 65536;
int a2a3 = (a2 + a3) * 65536;
int b1b2 = (b1 + b2) * 65536;
int bufferSizePerLine = (Width > Height ? Width : Height) * Channels;
unsigned char * bufferPerLine = (unsigned char*)malloc(bufferSizePerLine);
unsigned char * cacheData = (unsigned char*)malloc(Height * Width * Channels);
int WidthStep = Width * Channels;
for (int y = 0; y < Height; ++y) {
unsigned char * pRowInitial = inputBuffer + WidthStep * y;
unsigned char * pColumnInitial = cacheData + y * Channels;
gaussianHorizontal(bufferPerLine, pRowInitial, pColumnInitial, Width, Height, Channels, Width, a0a1, a2a3, b1b2, icprev, icnext);
}
int HeightStep = Height*Channels;
for (int x = 0; x < Width; ++x) {
unsigned char * pColInitial = outputBuffer + x*Channels;
unsigned char * pRowInitial = cacheData + HeightStep * x;
gaussianVertical(bufferPerLine, pRowInitial, pColInitial, Height, Width, Channels, a0a1, a2a3, b1b2, icprev, icnext);
}
free(bufferPerLine);
free(cacheData);
}
调用方法:
GaussianBlurFilter(输入图像数据,输出图像数据,宽度,高度,通道数,强度)
注:支持通道数分别为 1 ,3 ,4.
关于IIR相关知识,参阅 百度词条 "IIR数字滤波器"
http://baike.baidu.com/view/3088994.htm
天下武功，唯快不破。
本文只是抛砖引玉一下，若有其他相关问题或者需求也可以邮件联系俺探讨。
邮箱地址是:
gaozhihan@vip.qq.com
题外话:
很多网友一直推崇使用opencv,opencv的确十分强大,但是若是想要有更大的发展空间以及创造力.
还是要一步一个脚印去实现一些最基本的算法,扎实的基础才是构建上层建筑的基本条件.
俺目前只是把opencv当资料库来看,并不认为opencv可以用于绝大多数的商业项目.
若本文帮到您,厚颜无耻求微信扫码打个赏.
半径无关单核单线程最快速高斯模糊实现(附完整C代码)的更多相关文章
- 半径无关快速高斯模糊实现(附完整C代码)
之前,俺也发过不少快速高斯模糊算法. 俺一般认为,只要处理一千六百万像素彩色图片,在2.2GHz的CPU上单核单线程超过1秒的算法,都是不快的. 之前发的几个算法,在俺2.2GHz的CPU上耗时都会超 ...
- 快速双边滤波 附完整C代码
很早之前写过<双边滤波算法的简易实现bilateralFilter>. 当时学习参考的代码来自cuda的样例. 相关代码可以参阅: https://github.com/johng12/c ...
- 传统高斯模糊与优化算法(附完整C++代码)
高斯模糊(英语:Gaussian Blur),也叫高斯平滑,是在Adobe Photoshop.GIMP以及Paint.NET等图像处理软件中广泛使用的处理效果,通常用它来减少图像噪声以及降低细节层次 ...
- 【如何快速的开发一个完整的iOS直播app】(美颜篇)
原文转自:袁峥Seemygo 感谢分享.自我学习 前言 在看这篇之前,如果您还不了解直播原理,请查看这篇文章如何快速的开发一个完整的iOS直播app(原理篇) 开发一款直播app,美颜功能是很重 ...
- 【如何快速的开发一个完整的 iOS 直播 app】(美颜篇)
来源:袁峥Seemygo 链接:http://www.jianshu.com/p/4646894245ba 前言 在看这篇之前,如果您还不了解直播原理,请查看这篇文章如何快速的开发一个完整的iOS直播 ...
- 如何快速的开发一个完整的iOS直播app(美颜篇)
前言 在看这篇之前,如果您还不了解直播原理,请查看这篇文章如何快速的开发一个完整的iOS直播app(原理篇) 开发一款直播app,美颜功能是很重要的,如果没有美颜功能,可能分分钟钟掉粉千万,本篇主要讲 ...
- 【如何快速的开发一个完整的iOS直播app】(采集篇)
原文转自:袁峥Seemygo 感谢分享.自我学习 前言 在看这篇之前,如果您还不了解直播原理,请查看这篇文章如何快速的开发一个完整的iOS直播app(原理篇) 开发一款直播app,首先需要采集主 ...
- 【如何快速的开发一个完整的iOS直播app】(播放篇)
原文转自:袁峥Seemygo 感谢分享.自我学习 前言 在看这篇之前,如果您还不了解直播原理,请查看上篇文章如何快速的开发一个完整的iOS直播app(原理篇) 开发一款直播app,集成ijkpl ...
- 【如何快速的开发一个完整的iOS直播app】(原理篇)
原文转自:袁峥Seemygo 感谢分享.自我学习 目录 [如何快速的开发一个完整的iOS直播app](原理篇) [如何快速的开发一个完整的iOS直播app](播放篇) [如何快速的开发一个完整的 ...
随机推荐
- 关闭Eclipse中自动弹出console的功能
当我们使用Eclipse编写代码的时候,一般都会使界面最大化,如果这时tomcat服务器处在运行的状态,那么当后台有打印内容,比如日志输出,代码程序报错输出时,console就会自动弹出,很不方便. ...
- activiti 工作流
1. 工作流的概念 工作流(Workflow),就是“业务过程的部分或整体在计算机应用环境下的自动化”,它主要解决的是“使在多个参与者之间按照某种预定义的规则传递文档.信息或任务的过程自动进行,从而实 ...
- CentOS 7.2 搭建 Openvpn 服务器
本文将以目前最新的openvpn-2.3.13.tar.gz为例来介绍如何在Linux系统中安装.配置及使用OpenVPN. 在这里,我们选用了一台预装CentOS 7.2 64位系统的计算机作为Op ...
- java线程控制、状态同步、volatile、Thread.interupt以及ConcurrentLinkedQueue
在有些严格的系统中,我们需要做到干净的停止线程并清理相关状态.涉及到这个主题会带出很多的相关点,简单的总结如下: 我们知道,在java中,有一个volatile关键字,其官方说明(https://do ...
- mysql max_allowed_packet过小导致的prepare失败
最近公司一台阿里云上模拟环境突然好好地就出错了额,总提示:"Unknown prepared statement handler (stmt) given to DEALLOCATE PRE ...
- C# Out 传值
public void Out(out int a, out int b) {//out相当于return返回值 //可以返回多个值 //拿过来变量名的时候,里面默认为空值 a=1; b=2; } s ...
- C#加密算法总结
C#加密算法总结 MD5加密 /// <summary> /// MD5加密 /// </summary> /// <param name="strPwd&qu ...
- 实验12：Problem C: 重载字符的加减法
Home Web Board ProblemSet Standing Status Statistics Problem C: 重载字符的加减法 Problem C: 重载字符的加减法 Time ...
- SharePoint 2013 JQuery Asset Picket
var b = new AssetPickerConfig(""); b.ClientID = ""; b.DefaultAssetLocation = & ...
- 使用Kindle4rss推送自己感兴趣的博文
微信是个好东西,信息量超大,正能量的东西居多,但信息过载的滋味也很不好受,浏览了一大堆铺天盖地的信息后,关上手机后大脑又重新回到空白.所以还是喜欢用RSS聚合功能,自己去订阅优秀的博客或新闻,当有更新 ...