Opencv's understanding of depth, dims, channels, step, Data, elemsize, and data address computing in mat

Http://ggicci.blog.163.com/blog/static/210364096201261052543349/

Title:

• CV: mat
• Depth/dims/channels/step/data/elemsize

• The class mat represents an N-dimen=dense numerical single-channel or multi-channel array. it can be used to store (MAT class objects are used to represent a multi-dimensional single-channel or multi-channel dense array, which can be used to store the following things)
  • Real or complex-valued vectors or matrices (real or composite value vectors and matrices)
  • Grayscale or color images (grayscale or color)
  • Voxel volumes (Stereo element)
  • Vector Fields)
  • Point Clouds)
  • Tensors (Tensor)
  • Histograms (though, very high-dimen1_histograms may be better stored in a sparsemat) (histogram, preferably stored in sparsemat at high latitudes)

• The C struct in the old version of opencv contains cvmat and cvmatnd. Currently, I use Version 2.3. The document in it indicates that cvmat and cvmatnd are discarded and replaced by mat in the C ++ encapsulation, in addition, the old version also has an iplimage, which is also replaced by mat (refer to the structure, class, and emgu in blog opencv. the table corresponding to the CV ).

• Formula for Calculating the address of data elements in a matrix (m:

ADDR (M_{I0, I1 ,... IM-1)
= M. Data + M. Step [0] * I0 + M. Step [1] * I1 +... + M. Step M-1] * IM-1 (M = M. dims M dimension)}

• Data: A pointer in the mat object pointing to a memory (uchar * data) in which the matrix data is stored)
• Dims: the dimension of the matrix represented by mat. For example, a matrix of 3*4 is two dimensions, and a matrix of 3*4*5 is three dimensions.
• Channels: channel. The number of values owned by each matrix element in the matrix. For example, there are 12 elements in the 3*4 matrix. If each element has three values, in this case, the matrix is 3 channels, that is, channels = 3. Generally, a color image has three channels: Red, green, and blue.
• Depth: depth, that is, the bits of each pixel, In the mat of opencv. in depth (), A number ranging from 0 to 6 represents different digits: Enum {cv_8u = 0, cv_8s = 1,
  Cv_16u = 2, cv_16s = 3, cv_32s = 4, cv_32f = 5, cv_64f = 6}; visible 0 and 1 represent 8 bits, 2 and 3 represent 16 bits, 4 and 5 represent 32 bits, and 6 represent 64 bits;
• Step: it is an array that defines the layout of the matrix. For details, see the image analysis below. Also, pay attention to Step 1 (STEP/elemsize1), M. step M-1] is always equal to elemsize, M. step 1 (S-1) is always equal to channels;
• Elemsize: The data size of each element in the Matrix. If the data type in mat is cv_8u, elemsize = 1, cv_8uc3, elemsize = 3, cv_16uc2, and elemsize = 4; remember, another elemsize1 represents the size of the Data Type in the matrix, that is, elemsize/channels.
  Size

Image Analysis 1: Two-Dimensional storage (stored row by row)

The above is a matrix of 3x4. Assume that the data type is cv_8u, that is, the uchar type of a single channel.

• This is a two-dimensional matrix, so the dimension is 2 (M. dims = 2 );
• M. rows = 3; M. Cols = 4;
• Sizeof (uchar) = 1, then the size of each data element is 1 (M. elemsize () = 1, M. elemsize1 () = 1 );
• Cv_8u to obtain M. Depth () = 0, M. channels () = 1;
• Because it is a two-dimensional matrix, the step array has only two values. Step [0] And step [1] represent the data size of a row and the data size of an element respectively, then M. step [0] = 4, M. step [1] = 1;
• M. Step1 (0) = M. Cols = 4; M. Step1 (1) = 1;

Assume that the above matrix data type is cv_8uc3, that is, three channels

• M. dims = 2; M. channels () = 3; M. Depth () = 0;
• M. elemsize () = 3 (each element contains 3 uchar values) M. elemsize1 () = 1 (elemsize/channels)
• M. step [0] = m. cols * m. elemsize () = 12, M. step [1] = m. channels () * m. elemsize1 () = m. elemsize () = 3;
• M. Step (0) = M. Cols * M. channels () = 12; M. Step (1) = M. channels () = 3;

Image Analysis 2: 3D storage (stored plane by plane)

The above is a matrix of 3x4x6. Assume that the data type is cv_16sc4, that is, the short type.

• M. dims = 3; M. channels () = 4; M. elemsize1 () = sizeof (short) = 2;
• M. rows = M. Cols =-1;
• M. elemsize () = m. elemsize1 () * m. channels () = m. step [M. dims-1] = m. step [2] = 2*4 = 8;
• M. Step [0] = 4*6 * M. elemsize () = 192;
• M. Step [1] = 6 * M. elemsize () = 48;
• M. Step [2] = M. elemsize () = 8;
• M. Step1 (0) = M. Step [0]/M. elemsize () = 48/2 = 96 (first dimension (number of elements) * number of channels );
• M. step 1 (1) = m. step [1]/m. elemsize () = 12/2 = 24 (second dimension (number of elements in the row/column width) * number of channels );
• M. step 1 (2) = m. step [2]/m. elemsize () = m. channels () = 4 (third dimension (element) * number of channels );