In breast diffuse optical tomography, the introduction of the L1-norm regularization greatly improves the quality of the reconstructed image. However, the non-differentiable property of the objective function leads to exceeding difficulty in the optimization process. A new reconstruction method based on the L1-norm regularization with the non-negative restriction is proposed. To easily solve the first-order gradient of the objective function, the non-negative prior information is introduced. The optimization process is then well simplified and accelerated. Both the numerical simulations and the phantom experiments demonstrate that this new method can obtain much better results than the conventional regularization methods, and its process is more simple and faster.