How to solve 2d fused lasso in cvx

minimize( 0.5sum(square(a-theta))+lambdanorm(D*theta,1) )
D=
the origin image gets more fuzzy how to solve this problem

This is not a modeling forum. if you have a well-defined convex optimization problem and need help implementing it in CVX, please ask here. For help in image processing model formulation, you need to look elsewhere.