This paper presents a proposal of depth estimation method which employs empirical modeling of cost function based on Maximization of A posteriori Probability (MAP) rule. The proposed method allows for unsupervised depth estimation without a need for usage of arbitrary settings or control parameters, like Smoothing Coefficient in Depth Estimation Reference Software (DERS), which was used as a reference. The attained quality of generated depth maps is comparable to a case when supervised depth estimation is used, and such parameters are manually optimized. In the case when sub-optimal settings of control parameters in supervised depth estimation with DERS is used, the proposed method provides gains of about 2.8dB measured in average PSNR quality of virtual views synthesized with the use of estimated depth maps in the tested sequence set. 1