OpenCV2.1之PCA、CvvImage、IplImage和CvMat的相互转化

1、PCA Class Introduction----C++ 

class PCA

{

public:

    // default constructor

    PCA();

    // computes PCA for a set of vectors stored as data rows or columns.

    PCA(const Mat& data, const Mat& mean, int flags, int maxComponents=0);

    // computes PCA for a set of vectors stored as data rows or columns

    PCA& operator()(const Mat& data, const Mat& mean, int flags, int maxComponents=0);

    // projects vector into the principal components space

    Mat project(const Mat& vec) const;

    void project(const Mat& vec, Mat& result) const;

    // reconstructs the vector from its PC projection

    Mat backProject(const Mat& vec) const;

    void backProject(const Mat& vec, Mat& result) const; 

    // eigenvectors of the PC space, stored as the matrix rows

    Mat eigenvectors;

    // the corresponding eigenvalues; not used for PCA compression/decompression

    Mat eigenvalues;

    // mean vector, subtracted from the projected vector

    // or added to the reconstructed vector

    Mat mean;

};

 注：PCA constructors 补充说明

PCA::PCA(const Mat& data, const Mat& mean, int flags, int maxComponents=0)

 Parameters:   

l  data – the input samples, stored as the matrix rows or as the matrix columns

l  mean – the optional mean value. If the matrix is empty ( Mat() ), the mean is computed from the data.

l  flags –

operation flags. Currently the parameter is only used to specify the data layout.

ü  CV_PCA_DATA_AS_ROWS Indicates that the input samples are stored as matrix rows.

ü  CV_PCA_DATA_AS_COLS Indicates that the input samples are stored as matrix columns.

l  maxComponents – The maximum number of components that PCA should retain. By default, all the components are retained. 

  

PCA Class Test

PCA compressPCA(const Mat& pcaset, int maxComponents, const Mat& testset, Mat& compressed)

{

    PCA pca(pcaset, // pass the data

            Mat(),// we do not have a pre-computed mean vector,

                   // so let the PCA engine to compute it

            CV_PCA_DATA_AS_ROW,// indicate that the vectors

                               // are stored as matrix rows

                               // (use CV_PCA_DATA_AS_COL if the vectors are

                               // the matrix columns)

            maxComponents// specify, how many principal components to retain

            );

    // if there is no test data, just return the computed basis, ready-to-use

    if( !testset.data )

        return pca;

    CV_Assert( testset.cols == pcaset.cols ); 

    compressed.create(testset.rows, maxComponents, testset.type()); 

    Mat reconstructed;

    for( int i = 0; i < testset.rows; i++ )

    {

        Mat vec = testset.row(i), coeffs = compressed.row(i);

        // compress the vector, the result will be stored

        // in the i-th row of the output matrix

        pca.project(vec, coeffs);

        // and then reconstruct it

        pca.backProject(coeffs, reconstructed);

        // and measure the error

        printf("

    }

    return pca;

}

2、 PCA

          //初始化数据
        CvMat* pData = cvCreateMat( 总的样本数, 每个样本的维数, CV_32FC1 );  //每一行表示一个样本
        CvMat* pMean = cvCreateMat(1, 样本的维数, CV_32FC1);
        CvMat* pEigVals = cvCreateMat(1, 样本的维数, CV_32FC1); //pEigVals中的每个数表示一个特征值
        CvMat* pEigVecs = cvCreateMat( 样本的维数, 样本的维数, CV_32FC1);  //每一行表示一个特征向量

 

        //PCA处理,计算出平均向量pMean,特征值pEigVals和特征向量pEigVecs
        cvCalcPCA( pData, pMean, pEigVals, pEigVecs, CV_PCA_DATA_AS_ROW );

 

        //选出前P个特征向量(主成份),然后投影,结果保存在pResult中            
        CvMat* pResult = cvCreateMat( 总的样本数, PCA变换后的样本维数(即主成份的数目), CV_32FC1 );
        cvProjectPCA( pData, pMean, pEigVecs, pResult );

 

        //重构,结果保存在pRecon中
        CvMat* pRecon = cvCreateMat( 总的样本数, 每个样本的维数, CV_32FC1 );
        cvBackProjectPCA( pResult, pMean, pEigVecs, pRecon );

 

        //重构误差的计算
        //计算pRecon和pData的"差"就可以了.

 

        附：读取矩阵pData中第i行的第j个元素的方法：cvmGet( pData, i, j ).

 

3、CvvImage

      利用CvvImage可以方便地在MFC中画图（DrawToHdc函数），但是在调用Copyof这个函数时，要写成

      CvvImage img;

      IplImage src;

      img.Copyof( src, src->nChannels),

      否则，如果写成img.Copyof(src),src是单通道的话，会出现错误!

      CvvImage不仅能处理3通道，也能处理单通道。

4、IplImage和CvMat的相互转化

     IplImage* img = cvLoadImage( "d:\\1.jpg" ); 
     CvMat matHearder;
     CvMat* mat = cvGetMat( img, &matHearder );

 

      int i;
      for( i = 0; i < 10; i++  )

           printf( "%d  ", (u_char)img->imageData[i] );

      printf( "\n" );

 

      for( i = 0; i < 10; i++  )

           printf( "%d  ", mat->data.ptr[i] );

     注意：（1）有些函数并不会自动将IplImage转化为CvMat,如pca和svd的相关函数。

             （2）上述转换后mat中元素的范围为[0,255],img的范围是[-128,127].

             （3）要把img中的数据转换成无符号整数时，需要强制类型转换 u_char;