局部二值模式LBP（Local Binary Pattern）实现代码

            局部二值模式LBP由于其简单、高效在目标检测、目标识别、图像检索等领域得到了广泛的应用。现如今，LBP的变体不下上十种，如VLBP、SILBP等。要了解更多关于LBP的知识可以参考：纹理分类-全局特征LBP及相关文献。

          下面介绍的三种LBP算子的区别见下图：

            本文不介绍LBP的理论知识，只是简单的说说LBP及其变体的实现：

            1）最原始的LBP算子的实现代码           

void lbp( Mat& _src, Mat& _dst) 
{	
	if ( -_dst.empty())
		_dst.create(_src.rows, _src.cols, CV_8UC1);
	_dst = cv::Scalar::all(0);
	// calculate patterns
	for(int i=1;i<_src.rows-1;i++)
	{
		for(int j=1;j<_src.cols-1;j++) 
		{
			uchar center = _src.at<uchar>(i,j);
			unsigned char code = 0;
			code |= (_src.at<uchar>(i-1,j-1) >= center) << 7;
			code |= (_src.at<uchar>(i-1,j) >= center) << 6;
			code |= (_src.at<uchar>(i-1,j+1) >= center) << 5;
			code |= (_src.at<uchar>(i,j+1) >= center) << 4;
			code |= (_src.at<uchar>(i+1,j+1) >= center) << 3;
			code |= (_src.at<uchar>(i+1,j) >= center) << 2;
			code |= (_src.at<uchar>(i+1,j-1) >= center) << 1;
			code |= (_src.at<uchar>(i,j-1) >= center) << 0;
			_dst.at<unsigned char>(i,j) = code;
		}
	}
}

         2）带有容忍银子的LBP算子

void lbp_with_factor( Mat& _src, Mat& _dst, uchar factor) // factor default 3
{	
	if ( -_dst.empty())
		_dst.create(_src.rows, _src.cols, CV_8UC1);
	_dst = cv::Scalar::all(0);
	// calculate patterns
	for(int i=1;i<_src.rows-1;i++)
	{
		for(int j=1;j<_src.cols-1;j++) 
		{
			uchar center = _src.at<uchar>(i,j)+factor;
			unsigned char code = 0;
			code |= (_src.at<uchar>(i-1,j-1) >= center) << 7;
			code |= (_src.at<uchar>(i-1,j) >= center) << 6;
			code |= (_src.at<uchar>(i-1,j+1) >= center) << 5;
			code |= (_src.at<uchar>(i,j+1) >= center) << 4;
			code |= (_src.at<uchar>(i+1,j+1) >= center) << 3;
			code |= (_src.at<uchar>(i+1,j) >= center) << 2;
			code |= (_src.at<uchar>(i+1,j-1) >= center) << 1;
			code |= (_src.at<uchar>(i,j-1) >= center) << 0; 
			_dst.at<unsigned char>(i,j) = code;
		}
	}
}

          3）带有乘性因子的LBP算子

// 这里需要采用2个字符来编码，如00、01、10、11
void lbp_with_factor( Mat& _src, Mat& _dst, float factor)  // facotr default 0.1
{	
	if ( -_dst.empty())
		_dst.create(_src.rows, _src.cols, CV_8UC1);
	_dst = cv::Scalar::all(0);
	// calculate patterns
	for(int i=1;i<_src.rows-1;i++)
	{
		for(int j=1;j<_src.cols-1;j++) 
		{
			uchar center = _src.at<uchar>(i,j);
			int up = center*(1.0+factor);
			int down = center*(1.0+factor);
			unsigned char code = 0;
			/*code |= (_src.at<uchar>(i-1,j-1) >= center ) << 7;
			code |= (_src.at<uchar>(i-1,j) >= center) << 6;
			code |= (_src.at<uchar>(i-1,j+1) >= center) << 5;
			code |= (_src.at<uchar>(i,j+1) >= center) << 4;
			code |= (_src.at<uchar>(i+1,j+1) >= center) << 3;
			code |= (_src.at<uchar>(i+1,j) >= center) << 2;
			code |= (_src.at<uchar>(i+1,j-1) >= center) << 1;
			code |= (_src.at<uchar>(i,j-1) >= center) << 0; 
			_dst.at<unsigned char>(i,j) = code;*/
		}
	}
}

        欲了解更多参考论文 ：Modeling Pixel Process with Scale Invariant Local Patterns for Background Subtraction in Complex Scenes。