dummy

1 Mode

Below the output of three modes:

Filter size must be odd number, eg: 3x3, 5x5. here we consider the situation which the size of image is not less than the size of filter.

let image: m x m and filter: n x n

mode size of ouptut extend matrix size
full m + n - 1 m + 2(n - 1)
same m m + n - 1
valid m - n + 1 m

if: A is 6x6, m = 6 V is 3x3, n = 3

1.1 full

mode size of ouptut extend matrix size
full 6+3-1 = 8 6+2*(3-1) = 10

1.2 same

mode size of ouptut extend matrix size
same 6 6+3-1 = 8

1.3 valid

mode size of ouptut extend matrix size
valid 6-3+1 = 4 6

2 Codes

Only using numpy implement the conv2d, and compare the ouput with scipy.signal.convolve2d.

3 References

  1. http://www.songho.ca/dsp/convolution/convolution2d_example.html?source=post_page

  2. https://blog.csdn.net/leviopku/article/details/80327478

  3. https://docs.scipy.org/doc/numpy/reference/generated/numpy.convolve.html

  4. https://blog.csdn.net/majinlei121/article/details/50256049