7_Kronecker

7_Kronecker - EE 211A Fall Quarter, 2011 Instructor: John...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: EE 211A Fall Quarter, 2011 Instructor: John Villasenor Digital Image Processing I Handout 7 Separability and Kronecker Products: An Example with N = 2 Let v and u be 2 × 2 matrices, where ⎡ྎv(0,0) v(0,1)⎤ྏ ⎡ྎu (0,0) u (0,1)⎤ྏ v = ⎢ྎ ⎥ྏ, u = ⎢ྎ u (1,0) u (1,1) ⎥ྏ ⎣ྏ v(1,0) v(1,1) ⎦ྏ ⎣ྏ ⎦ྏ 1 1 and v(k , l ) = ∑ m=0 ∑ n=0 u(m, n)akl (m, n). If v ʹȃ and u ʹȃ are row-ordered versions of v and u , then we can write ⎡ྎv(0,0)⎤ྏ ⎡ྎa00 (0,0) ⎢ྎ v(0,1) ⎥ྏ ⎢ྎ a (0,0) ⎥ྏ = ⎢ྎ 01 v ʹȃ = ⎢ྎ ⎢ྎ v(1,0) ⎥ྏ ⎢ྎ a10 (0,0) ⎢ྎ ⎥ྏ ⎢ྎ ⎣ྏ v(1,1) ⎦ྏ ⎣ྏ a11 (0,0) a00 (0,1) a00 (1,0) a00 (1,1)⎤ྏ ⎡ྎu (0,0)⎤ྏ a01 (0,1) a01 (1,0) a01 (1,1) ⎥ྏ ⎢ྎ u (0,1) ⎥ྏ ⎥ྏ ⋅ ⎢ྎ ⎥ྏ = au ʹȃ. a10 (0,1) a10 (1,0) a10 (1,1) ⎥ྏ ⎢ྎ u (1,0) ⎥ྏ ⎥ྏ ⎢ྎ ⎥ྏ a11 (0,1) a11 (1,0) a11 (1,1) ⎦ྏ ⎣ྏ u (1,1) ⎦ྏ In the general case, all elements akl (m, n) are independent, and no further simplification is possible. Calculation of v ʹȃ costs N 4 = 16 multiplies. Now suppose that akl (m, n) = b(k , m)c(l , n).The above matrix product is: ⎡ྎv(0, 0) ⎤ྏ ⎡ྎb(0, 0)c(0, 0) b(0, 0)c(0,1) b(0,1)c(0, 0) b(0,1)c(0,1) ⎤ྏ ⎡ྎu (0, 0) ⎤ྏ ⎢ྎ v(0,1) ⎥ྏ ⎢ྎ b(0, 0)c(1, 0) b(0, 0)c(1,1) b(0,1)c(1, 0) b(0,1)c(1,1) ⎥ྏ ⎢ྎ u (0,1) ⎥ྏ ⎥ྏ = ⎢ྎ ⎥ྏ ⋅ ⎢ྎ ⎥ྏ = auʹȃ. vʹȃ = ⎢ྎ ⎢ྎ v(1, 0) ⎥ྏ ⎢ྎ b(1, 0)c(0, 0) b(1, 0)c(0,1) b(1,1)c(0, 0) b(1,1)c (0,1) ⎥ྏ ⎢ྎ u (1, 0) ⎥ྏ ⎢ྎ ⎥ྏ ⎢ྎ ⎥ྏ ⎢ྎ ⎥ྏ ⎣ྏ v(1,1) ⎦ྏ ⎣ྏ b(1, 0)c(1, 0) b(1, 0)c(1,1) b(1,1)c(1, 0) b(1,1)c(1,1) ⎦ྏ ⎣ྏ u (1,1) ⎦ྏ The matrix a can be expressed as a Kronecker product a = b ⊗ c. ⎡ྎb(0,0)c b(0,1)c ⎤ྏ ⇒ a = ⎢ྎ ⎥ྏ. ⎣ྏ b(1,0)c b(1,1)c ⎦ྏ In addition, v can be expressed: ⎡ྎb(0,0) b(0,1)⎤ྏ ⎡ྎu (0,0) u (0,1)⎤ྏ ⎡ྎc(0,0) c(1,0)⎤ྏ T v = bu c = ⎢ྎ ⎥ྏ ⋅ ⎢ྎ ⎥ྏ ⋅ ⎢ྎ ⎥ྏ ⎣ྏ b(1,0) b(1,1) ⎦ྏ ⎣ྏ u (1,0) u (1,1) ⎦ྏ ⎣ྏ c(0,1) c(1,1) ⎦ྏ 1 Verify: ⎡ྎb(0,0) b(0,1)⎤ྏ ⎡ྎc(0,0)u (0,0) + c(0,1)u (0,1) c(1,0)u (0,0) + c(1,1)u (0,1)⎤ྏ v = ⎢ྎ ⎥ྏ ⋅ ⎢ྎ ⎥ྏ ⎣ྏ b(1,0) b(1,1) ⎦ྏ ⎣ྏ c(0,0)u (1,0) + c(0,1)u (1,1) c(1,0)u (1,0) + c(1,1)u (1,1) ⎦ྏ ⇒ v(0,0) = b(0,0)[c(0,0)u (0,0) + c(0,1)u (0,1)] + b(0,1)[c(0,0)u (1,0) + c(0,1)u (1,1)], and so on. On an element by element basis, we can see that if akl (m, n) is separable, where a = b ⊗ c, T then v ʹȃ = a u ʹȃ can be expressed as = c . v b u N 2 ×1 N 2 ×N 2 N 2 ×1 N ×N The total multiplies for the separable case is 2 N 3 . 2 N×N N ×N N ×N ...
View Full Document

This note was uploaded on 12/27/2011 for the course EE211A 211A taught by Professor Villasenor during the Spring '11 term at UCLA.

Ask a homework question - tutors are online