Home  |  About Us  |  Link To Us  |  FAQ  |  Contact

# Kernel decomposition 1.0

Date Added: May 30, 2013  |  Visits: 181

This function does the decomposition of a separable nD kernel intoits 1D components, such that a convolution with each of thesecomponents yields the same result as a convolution with the full nDkernel, at a drastic reduction in computational cost.SYNTAX:======= [K1,KN,ERR] = DECOMPOSE_KERNEL(H)computes a set of 1D kernels K1{1}, K1{2}, ... K1{N} such that theconvolution of an image with all of these in turn yields the sameresult as a convolution with the N-dimensional input kernel H: RES1 = CONVN(IMG,H); RES2 = IMG; FOR II=1:LENGTH(K1) RES2 = CONVN(RES2,K1{II}); ENDKN is the reconstruction of the original kernel H from the 1Dkernels K1, and ERR is the sum of absolute differences between Hand KN.The syntax mimics Dirk-Jan Kroon's submission to the FileExchange(see below).EXPLANATION:============In general, for a 2D kernel H, the convolution with 2D image F: G = F * His identical to the convolution of the image with column vector H1and convolution of the result with row vector H2: G = ( F * H1 ) * H2 .In MATLAB speak, this means that> CONV2(F,H) == CONV2(CONV2(F,H1),H2)Because of the properties of the convolution, ( F * H1 ) * H2 = F * ( H1 * H2 ) ,meaning that the convolution of the two 1D filters with each otherresults in the original filter H. And because H1 is a column vectorand H2 a row vector, H = H1 * H2 = H1 H2 .Thus, we need to find two vectors whose product yields the matrix H.In MATLAB speak we need to solve the equation> H1*H2 == HThe function in the standard MATLAB toolbox, FILTER2, does justthis, and it does it using singular value decomposition: U S V' = H , H1(i) = U(i,1) S(1,1)^0.5 , H2(i) = V(i,1)* S(1,1)^0.5 . (the * here is the conjugate!)Note that, if the kernel H is separable, all values of S are zeroexcept S(1,1). Also note that this is an under-determined problem,in the sense that H = H1 H2 = ( a H1 ) ( 1/a H2 ) ;that is, it is possible to multiply one 1D kernel with any valueand compensate by dividing the other kernel with the same value.Our solution will, in effect, just choose one of the infinite numberof (equivalent) solutions.To extend this concept to nD, what we need to understand is that itis possible to collapse all dimensions except one, obtaining a 2Dmatrix, and solve the above equation. This results in a 1D kerneland an (n-1)D kernel. Repeat the process until all you have is aset of 1D kernels and you're done!This function is inspired by a solution to this problem thatDirk-Jan Kroon posted on the File Exchange recently:http://www.mathworks.com/matlabcentral/fil...l-in-1d-kernelsHis solution does the whole decomposition in one go, by setting upone big set of equations. He noted a problem with negative values,which produce complex 1D kernels. The magnitude of the result iscorrect, but the sign is lost. He needs to resort to some heuristicto determine the sign of each element. What he didn't notice (ordidn't mention) is the problem that his solution has with 0 values.The SVD solution doesn't have this problem, although it sometimesdoes produce a slightly worse solution. For example, in the firstexample below, Dirk-Jan Kroon's solution is exact, whereas this oneproduces a very small error. Where Dirk-Jan Kroon's solution cannotfind the exact solution, this algorithm generally does better.EXAMPLES:=========Simplest 5D example: H = ones(5,7,4,1,5); [K1,~,err] = SeparateKernel(H); % D.Kroon's submission to FileEx. err [k1,~,err] = decompose_kernel(H); err2D example taken from Dirk-Jan Kroon's submission: a = permute(rand(4,1),[1 2 3])-0.5; b = permute(rand(4,1),[2 1 3])-0.5; H = repmat(a,[1 4]).*repmat(b,[4 1]); [K1,~,err] = SeparateKernel(H); err [k1,~,err] = decompose_kernel(H); err2D example for which Dirk-Jan Kroon's solution has problems: H = [1,2,3,2,1]'*[1,1,3,0,3,1,1]; [K1,~,err] = SeparateKernel(H); err [k1,~,err] = decompose_kernel(H); err3D example that's not separable: H = rand(5,5,3); [K1,~,err] = SeparateKernel(H); err [k1,~,err] = decompose_kernel(H);Example to apply a convolution using the decomposed kernel: img = randn(50,50,50); h = ones(7,7,7); tic; res1 = convn(img,h); toc k1 = decompose_kernel(h); tic; res2 = img; for ii=1:length(k1) res2 = convn(res2,k1{ii}); end toc rms_diff = sqrt(mean((res1(:)-res2(:)).^2))

 Requirements: No special requirements Platforms: Matlab Keyword: Complex,  Correct,  Determine,  Heuristic,  Magnitude,  Negative,  Produce,  Resort Users rating: 0/10