matlab-(含教程)基于全变量DAM算法的图像去噪matlab仿真

function [Trans, TransSizes,g_num] = decompose(m,n,TV,num) modd = rem(m,2); nodd = rem(n,2); % Set the linear transformations. if n==1 if strcmpi(TV, 'l1') if num<2 error 'For l1 TV num should be greater or equal to 2'; end g_num = num; TransSizes = 2*floor((m-(-(num-2):1)')/num); max_size = max(TransSizes); Trans = zeros(max_size,g_num); for i=1:g_num Trans(1:2:TransSizes(i),i) = i:num:((TransSizes(i)/2)*num-2+i); Trans(2:2:TransSizes(i),i) = (i:num:((TransSizes(i)/2)*num-2+i))+1; end else if num<3 error 'For iso TV num should be greater or equal to 3'; end g_num = num; TransSizes = 3*floor((m-(-(num-3):2)')/num); max_size = max(TransSizes); Trans = zeros(max_size,g_num); for i=1:g_num Trans(1:3:TransSizes(i),i) = i:num:((TransSizes(i)/3)*num-3+i); Trans(2:3:TransSizes(i),i) = (i:num:((TransSizes(i)/3)*num-3+i))+1; Trans(3:3:TransSizes(i),i) = (i:num:((TransSizes(i)/3)*num-3+i))+2; end end else %n>1 if strcmpi(TV, 'l1') g_num = 4; TransSizes = zeros(g_num,1); TransSizes(1:2) = 2*n*floor((m-(0:1)')/2); TransSizes(3:4) =2*m*floor((n-(0:1)')/2); max_size = max(TransSizes); Trans = zeros(max_size,g_num); Trans(1:TransSizes(1),1)= uint32(repmat((1:(m-modd))',n,1)+... kron((0:(n-1))',ones(m-modd,1))*m); Trans(1:TransSizes(2),2)= uint32(repmat((2:(m-1+modd))',n,1)+... kron((0:(n-1))',ones(m-2+modd,1))*m); Trans(1:TransSizes(3),3)= uint32(repmat((1:m:(m*(n-nodd)))',m,1)+... kron((0:(m-1))',ones(n-nodd,1))); Trans(1:TransSizes(4),4)= uint32(repmat(((m+1):m:(m*(n-1+nodd)))',m,1)+... kron((0:(m-1))',ones(n-2+nodd,1))); else % TV='iso' g_num = 3; TransSizes = zeros(g_num,2); NinRow = zeros(3); % Number of r-shaped triples in a column for each level of % diagonal (that might be cyclic) shift. NinCol = [floor((m+1)/3);floor(m/3);floor((m-1)/3)]; for i=1:3 % column shift = equivalent to change in % transformation index % Number of r-shaped triples in a rwo for each level of % diagonal (that might be cyclic) shift. % i=1 i=2 i=3 NinRow(:,i) = [floor((n+1-rem(i-1,3))/3);... % n+1 n n-1 m+1 floor((n+1-rem(i,3))/3);... % n n-1 n+1 m floor((n+1-rem(i+1,3))/3)]; % n-1 n+1 n m-1 TransSizes(i,1) = 3*NinCol'*NinRow(:,i); % Compute the length of % the transformed 1D % vector. TransSizes(i,2) = 2*(NinCol(rem(3+rem(n,3)-i,3)+1)+NinRow(rem(2+rem(m,3),3)+1,i)); end max_size = max(max(TransSizes)); Trans = zeros(max_size,g_num,2); % The iso TV is a sum of 3-dimensional vector norms. % Each such vector is defined by the r-shaped triple % (x_{i,j+1},x_{i,j},x_{i+1,1}) for each pair of indices i<m,j<n. % The following code produce a one to one transformation of the % entities from 2D to 1D and vise versa according to the type of % decomposition. Throughout of the documentation of this code the % first entity of the triple x_{i,j+1} will be referred as the % column shift entity (cs-entity), the middle one as the base % entity (b-entity) and the last one will be referred as the row % shift entity (rs-entity). for i=1:3 %column shift = equivalent to change in % transformation index. % Mark the indices that will contain the values of the % rs-entities in the transformed 1D vector. indx = zeros(max_size,1);indx(1:3:TransSizes(i))=1; indx=logical(indx); % Set the boundaries of the domain of each level of % diagonal shift. indx_sep = cumsum([1;3*NinCol.*NinRow(:,i)]); % rs-entity start index in the left most column for each % level of diagonal shift. csSiF = (1:3)'; csSiF = (rem(i+csSiF+1,3)+1)*m+csSiF; % rs-entity end index in the left most column for each % level of diagonal shift. csEiF = csSiF + 3*(NinCol-1); % b&cs-couple entities start index in the left most column % for each level of diagonal shift. brsSiF = (1:3)'; brsSiF = rem(1+brsSiF+i,3)*m+brsSiF; % b&cs-couple entities end index in the left most column % for each level of diagonal shift. bcsEi = brsSiF + 3*(NinCol-1); for j=1:3 %row shift curr_indx = (indx_sep(j):(indx_sep(j+1)-1))'; curr_indx = indx(curr_indx).*curr_indx; curr_indx = curr_indx(curr_indx>0); Trans(curr_indx,i,1) = ... repmat((csSiF(j):3:csEiF(j))',NinRow(j,i),1) + ... kron((0:3:(3*NinRow(j,i)-1))',ones(NinCol(j),1))*m; Trans(curr_indx+1,i,1) = ... repmat((brsSiF(j):3:bcsEi(j))',NinRow(j,i),1) + ... kron((0:3:(3*NinRow(j,i)-1))',ones(NinCol(j),1))*m; Trans(curr_indx+2,i,1) = ... repmat(1+(brsSiF(j):3:bcsEi(j))',NinRow(j,i),1) + ... kron((0:3:(3*NinRow(j,i)-1))',ones(NinCol(j),1))*m; end colIndx = rem(3+rem(n,3)-i,3)+1; brsSiL = m*(n-1)+colIndx; brsEiL = brsSiL+3*(NinCol(colIndx)-1); rowIndx = rem(2+rem(m,3),3)+1; bcsSiL = (rem(rowIndx+i-2,3)+1)*m; bcsEiL = bcsSiL+3*m*(NinRow(rowIndx,i)-1); Trans(1:2:TransSizes(i,2),i,2) = ... [(brsSiL:3:brsEiL)';(bcsSiL:3*m:bcsEiL)']; Trans(2:2:TransSizes(i,2),i,2) = ... [(brsSiL+1:3:brsEiL+1)';(bcsSiL+m:3*m:bcsEiL+m)']; end end end if num~=0 && (n>1) error('Not supported.'); end end