Coherent beam combining (CBC) with closely arranged centrosymmetric arrays is a promising way to obtain a high-brightness laser. An essential task in CBC is to actively control the piston phases of the input beams, maintaining the correct phasing to maximize the combination efficiency. By applying the neural network, the nonlinear mapping relationship between the far-field image and the piston phase could be established, so that the piston phase can be corrected quickly with one step, which caused widespread concern. However, there exists a piston-type phase ambiguity problem in the CBC system with centrosymmetric arrays, which means that multiple different piston phases may generate the same far-field image. This will prevent the far-field image from correctly reflecting the phase information, which will result in a performance degradation of the image-based intelligent algorithms. In this paper, we make a theoretical analysis of phase ambiguity. A method to solve phase ambiguity is proposed, which requires no additional optical devices. We designed simulations to verify our conclusions and methods. We believe that our work solves the phase ambiguity problem in theory and is conducive to improving the performance of image-based algorithms.