Improved Solution to the l0 Regularized Optimization Problem via Dictionary-Reduced Initial Guess