In recent years, there has been an increasing interest in diffuse optical tomography (DOT). The image accuracy and computational complexity are greatly affected by the reconstruction algorithm. This paper aims at providing an efficient reconstruction technique which can be implemented in VLSI for advanced portable devices. Various inhomogeneous absorption models are used in simulations of continuous-wave DOT systems. The accuracy and time consumption are calculated and compared for both frame and sub-frame modes. Mean square errors (MSE) are derived to quantify the accuracy of different modes. Decreasing the complexity of algorithm and enhancing the quality of image can be achieved by adjusting the truncated parameter and reconstructive mode. This study has shown that the sub-frame mode with truncated Jacobi Singular Value Decomposition (TJSVD) algorithm is the method with higher efficiency and better quality. It can also be implemented in VLSI for continuous-wave DOT.