Apply function to matrix blocks
Executes the function func on blocks of size block_size, taken from the matrix A. Both the matrix A, and the block can have any number of dimensions.
The different blocks are organized into a matrix, each block as a
single column, and passed as the first to the function handle
func. Any input arguments to colfilt after func
are passed to func after the blocks matrix.
Blocks can be of two different types as defined by the string block_type:
"distinct"Each block is completely distinct from the other, with no overlapping elements. func must return a matrix of exactly the same size as its input.
"sliding"Each possible block of size block_size inside A is used. func should act along the column dimension (be a column compression function) and return a vector of length equal to the number of columns of its input.
The optional argument subsize divides A into smaller pieces before generating the matrices with one block per column in order to save memory. It is currently only accepted for MATLAB compatibility.
If A is an indexed image, the second argument should be the
string "indexed" so that any required padding is done correctly.
The padding value will be 0 except for indexed images of class uint8
and uint16.
This function is mostly useful to apply moving or sliding window filter when block_type is "sliding". However, for many cases, specialized functions perform much faster. For the following common cases, consider the suggested alternatives;
A moving average filter is equivalent to convolve with a matrix
of 1/N sized block_size, where N is the total
number of elements in a block. Use
convn (A, (1/N) * ones (block_size) *, "same")
This is the equivalent to a dilation and erosion. Use imdilate or
imerode.
Same as dilation and erosion but with logical input. Use imdilate
or imerode with logical (A).
Use medfilt2 if A is only 2 dimensional, and ordfiltn
with the floor (prod (N/ 2) th element, where N is the
total number of elements in a block (add 1 if it is an even number).
Use ordfiltn.
Use stdfilt.
Use a matrix of 1 to perform convolution,
convn (A, ones (block_size), "same")
See also: bestblk, blockproc, col2im, im2col, nlfilter.
The following code
## Perform moving average filter with a 4x4 window A = magic (12) colfilt (A, [4 4], "sliding", @mean)
Produces the following output
A =
144 2 3 141 140 6 7 137 136 10 11 133
13 131 130 16 17 127 126 20 21 123 122 24
25 119 118 28 29 115 114 32 33 111 110 36
108 38 39 105 104 42 43 101 100 46 47 97
96 50 51 93 92 54 55 89 88 58 59 85
61 83 82 64 65 79 78 68 69 75 74 72
73 71 70 76 77 67 66 80 81 63 62 84
60 86 87 57 56 90 91 53 52 94 95 49
48 98 99 45 44 102 103 41 40 106 107 37
109 35 34 112 113 31 30 116 117 27 26 120
121 23 22 124 125 19 18 128 129 15 14 132
12 134 135 9 8 138 139 5 4 142 143 1
ans =
Columns 1 through 8:
42.812 54.375 54.625 54.375 54.125 54.375 54.625 54.375
54.375 72.500 72.500 72.500 72.500 72.500 72.500 72.500
57.375 72.500 72.500 72.500 72.500 72.500 72.500 72.500
54.375 72.500 72.500 72.500 72.500 72.500 72.500 72.500
51.375 72.500 72.500 72.500 72.500 72.500 72.500 72.500
54.375 72.500 72.500 72.500 72.500 72.500 72.500 72.500
57.375 72.500 72.500 72.500 72.500 72.500 72.500 72.500
54.375 72.500 72.500 72.500 72.500 72.500 72.500 72.500
51.375 72.500 72.500 72.500 72.500 72.500 72.500 72.500
54.375 72.500 72.500 72.500 72.500 72.500 72.500 72.500
39.062 54.375 54.625 54.375 54.125 54.375 54.625 54.375
27.938 36.250 36.250 36.250 36.250 36.250 36.250 36.250
Columns 9 through 12:
54.125 54.375 42.500 27.250
72.500 72.500 54.375 36.250
72.500 72.500 57.375 36.250
72.500 72.500 54.375 36.250
72.500 72.500 51.375 36.250
72.500 72.500 54.375 36.250
72.500 72.500 57.375 36.250
72.500 72.500 54.375 36.250
72.500 72.500 51.375 36.250
72.500 72.500 54.375 36.250
54.125 54.375 38.750 27.250
36.250 36.250 27.938 18.125
Package: image