Convolution

Separable Convolutions



• Break m?n kernel into m?1 row and 1?n column
  • Think of them as row, column vectors
• When possible, it reduces m*n multiplies (or accumulations) into m+n multiplies
• Implement in accum using two convolutions, m?1, then 1?n
• Use glSeparableFilter2D, provide row and column

Previous slide

Next slide

Back to first slide

View graphic version