cv::Mat与CvMat区别

用了OpenCV一段时间，发现有cv::Mat和CvMat都可以表示图像数据，只是却不明白他们之间到底什么区别，为什么一定要有这样的两种~Mat。后来跟踪查看了定义，才知道原来cv::Mat是一个类（Class），而CvMat是一个Struct，从类型上就可以知道二者存在很大区别。前者除了几个成员变量之外还有很多的成员函数和重载函数，可以实现很多的图像数据处理功能，而后者只有几个成员变量，要对其成员进行一些处理，需要借用别的函数。

typedef struct CvMat
{
    int type;
    int step;

    /* for internal use only */
    int* refcount;
    int hdr_refcount;

    union                                               //数据的指针
    {
        uchar* ptr;
        short* s;
        int* i;
        float* fl;
        double* db;
    } data;

#ifdef __cplusplus   
    union
    {
        int rows;
        int height;
    };

    union
    {
        int cols;
        int width;
    };
#else
    int rows;
    int cols;
#endif

}
CvMat;

 

CvMat中的data数据只是矩阵数据的首地址，分配的内存大小为行列乘积。对该内存块的操作需要调用函数，也可以使用指针索引。

 

 

class CV_EXPORTS Mat
{
public:
    // constructors
    Mat();
    // constructs matrix of the specified size and type
    // (_type is CV_8UC1, CV_64FC3, CV_32SC(12) etc.)
    Mat(int _rows, int _cols, int _type);
    Mat(Size _size, int _type);
    // constucts matrix and fills it with the specified value _s.
    Mat(int _rows, int _cols, int _type, const Scalar& _s);
    Mat(Size _size, int _type, const Scalar& _s);
    // copy constructor
    Mat(const Mat& m);
    // constructor for matrix headers pointing to user-allocated data
    Mat(int _rows, int _cols, int _type, void* _data, size_t _step=AUTO_STEP);
    Mat(Size _size, int _type, void* _data, size_t _step=AUTO_STEP);
    // creates a matrix header for a part of the bigger matrix
    Mat(const Mat& m, const Range& rowRange, const Range& colRange);
    Mat(const Mat& m, const Rect& roi);
    // converts old-style CvMat to the new matrix; the data is not copied by default
    Mat(const CvMat* m, bool copyData=false);
    // converts old-style IplImage to the new matrix; the data is not copied by default
    Mat(const IplImage* img, bool copyData=false);
    // builds matrix from std::vector with or without copying the data
    template<typename _Tp> Mat(const vector<_Tp>& vec, bool copyData=false);
    // helper constructor to compile matrix expressions
    Mat(const MatExpr_Base& expr);
    // destructor - calls release()
    ~Mat();
    // assignment operators
    Mat& operator = (const Mat& m);
    Mat& operator = (const MatExpr_Base& expr);

 

    operator MatExpr_<Mat, Mat>() const;

 

    // returns a new matrix header for the specified row
    Mat row(int y) const;
    // returns a new matrix header for the specified column
    Mat col(int x) const;
    // ... for the specified row span
    Mat rowRange(int startrow, int endrow) const;
    Mat rowRange(const Range& r) const;
    // ... for the specified column span
    Mat colRange(int startcol, int endcol) const;
    Mat colRange(const Range& r) const;
    // ... for the specified diagonal
    // (d=0 - the main diagonal,
    //  >0 - a diagonal from the lower half,
    //  <0 - a diagonal from the upper half)
    Mat diag(int d=0) const;
    // constructs a square diagonal matrix which main diagonal is vector "d"
    static Mat diag(const Mat& d);

 

    // returns deep copy of the matrix, i.e. the data is copied
    Mat clone() const;
    // copies the matrix content to "m".
    // It calls m.create(this->size(), this->type()).
    void copyTo( Mat& m ) const;
    // copies those matrix elements to "m" that are marked with non-zero mask elements.
    void copyTo( Mat& m, const Mat& mask ) const;
    // converts matrix to another datatype with optional scalng. See cvConvertScale.
    void convertTo( Mat& m, int rtype, double alpha=1, double beta=0 ) const;

 

    void assignTo( Mat& m, int type=-1 ) const;

 

    // sets every matrix element to s
    Mat& operator = (const Scalar& s);
    // sets some of the matrix elements to s, according to the mask
    Mat& setTo(const Scalar& s, const Mat& mask=Mat());
    // creates alternative matrix header for the same data, with different
    // number of channels and/or different number of rows. see cvReshape.
    Mat reshape(int _cn, int _rows=0) const;

 

    // matrix transposition by means of matrix expressions
    MatExpr_<MatExpr_Op2_<Mat, double, Mat, MatOp_T_<Mat> >, Mat>
    t() const;
    // matrix inversion by means of matrix expressions
    MatExpr_<MatExpr_Op2_<Mat, int, Mat, MatOp_Inv_<Mat> >, Mat>
        inv(int method=DECOMP_LU) const;
    MatExpr_<MatExpr_Op4_<Mat, Mat, double, char, Mat, MatOp_MulDiv_<Mat> >, Mat>
    // per-element matrix multiplication by means of matrix expressions
    mul(const Mat& m, double scale=1) const;
    MatExpr_<MatExpr_Op4_<Mat, Mat, double, char, Mat, MatOp_MulDiv_<Mat> >, Mat>
    mul(const MatExpr_<MatExpr_Op2_<Mat, double, Mat, MatOp_Scale_<Mat> >, Mat>& m, double scale=1) const;
    MatExpr_<MatExpr_Op4_<Mat, Mat, double, char, Mat, MatOp_MulDiv_<Mat> >, Mat>   
    mul(const MatExpr_<MatExpr_Op2_<Mat, double, Mat, MatOp_DivRS_<Mat> >, Mat>& m, double scale=1) const;
   
    // computes cross-product of 2 3D vectors
    Mat cross(const Mat& m) const;
    // computes dot-product
    double dot(const Mat& m) const;

 

    // Matlab-style matrix initialization
    static MatExpr_Initializer zeros(int rows, int cols, int type);
    static MatExpr_Initializer zeros(Size size, int type);
    static MatExpr_Initializer ones(int rows, int cols, int type);
    static MatExpr_Initializer ones(Size size, int type);
    static MatExpr_Initializer eye(int rows, int cols, int type);
    static MatExpr_Initializer eye(Size size, int type);

 

    // allocates new matrix data unless the matrix already has specified size and type.
    // previous data is unreferenced if needed.
    void create(int _rows, int _cols, int _type);
    void create(Size _size, int _type);
    // increases the reference counter; use with care to avoid memleaks
    void addref();
    // decreases reference counter;
    // deallocate the data when reference counter reaches 0.
    void release();

 

    // locates matrix header within a parent matrix. See below
    void locateROI( Size& wholeSize, Point& ofs ) const;
    // moves/resizes the current matrix ROI inside the parent matrix.
    Mat& adjustROI( int dtop, int dbottom, int dleft, int dright );
    // extracts a rectangular sub-matrix
    // (this is a generalized form of row, rowRange etc.)
    Mat operator()( Range rowRange, Range colRange ) const;
    Mat operator()( const Rect& roi ) const;

 

    // converts header to CvMat; no data is copied
    operator CvMat() const;
    // converts header to IplImage; no data is copied
    operator IplImage() const;
   
    // returns true iff the matrix data is continuous
    // (i.e. when there are no gaps between successive rows).
    // similar to CV_IS_MAT_CONT(cvmat->type)
    bool isContinuous() const;
    // returns element size in bytes,
    // similar to CV_ELEM_SIZE(cvmat->type)
    size_t elemSize() const;
    // returns the size of element channel in bytes.
    size_t elemSize1() const;
    // returns element type, similar to CV_MAT_TYPE(cvmat->type)
    int type() const;
    // returns element type, similar to CV_MAT_DEPTH(cvmat->type)
    int depth() const;
    // returns element type, similar to CV_MAT_CN(cvmat->type)
    int channels() const;
    // returns step/elemSize1()
    size_t step1() const;
    // returns matrix size:
    // width == number of columns, height == number of rows
    Size size() const;
    // returns true if matrix data is NULL
    bool empty() const;

 

    // returns pointer to y-th row
    uchar* ptr(int y=0);
    const uchar* ptr(int y=0) const;

 

    // template version of the above method
    template<typename _Tp> _Tp* ptr(int y=0);
    template<typename _Tp> const _Tp* ptr(int y=0) const;
   
    // template methods for read-write or read-only element access.
    // note that _Tp must match the actual matrix type -
    // the functions do not do any on-fly type conversion
    template<typename _Tp> _Tp& at(int y, int x);
    template<typename _Tp> _Tp& at(Point pt);
    template<typename _Tp> const _Tp& at(int y, int x) const;
    template<typename _Tp> const _Tp& at(Point pt) const;
   
    // template methods for iteration over matrix elements.
    // the iterators take care of skipping gaps in the end of rows (if any)
    template<typename _Tp> MatIterator_<_Tp> begin();
    template<typename _Tp> MatIterator_<_Tp> end();
    template<typename _Tp> MatConstIterator_<_Tp> begin() const;
    template<typename _Tp> MatConstIterator_<_Tp> end() const;

 

    enum { MAGIC_VAL=0x42FF0000, AUTO_STEP=0, CONTINUOUS_FLAG=CV_MAT_CONT_FLAG };

 

    // includes several bit-fields:
    //  * the magic signature
    //  * continuity flag
    //  * depth
    //  * number of channels
    int flags;
    // the number of rows and columns
    int rows, cols;
    // a distance between successive rows in bytes; includes the gap if any
    size_t step;
    // pointer to the data
    uchar* data;

 

    // pointer to the reference counter;
    // when matrix points to user-allocated data, the pointer is NULL
    int* refcount;
   
    // helper fields used in locateROI and adjustROI
    uchar* datastart;
    uchar* dataend;
};