This paper reviews the physical principles and information-theoretic description of ghost imaging, an imaging technique based on the high-order statistical correlation of optical fields. First, the physical principles of ghost imaging are introduced: the fundamental concepts of optical field correlation functions and first- and second-order coherence are expounded, and the high-order statistical correlation characteristics of typical light sources, including coherent states, thermal (pseudo-thermal) light, and entangled photon pairs, are analyzed. The imaging mechanism in which the degree of second-order coherence of the optical field exhibits a single-peaked distribution under both entangled photon-pair and (pseudo-)thermal light illumination, thereby establishing a point-to-point mapping between the object plane and the image plane, is discussed, together with the unified interpretation offered by the Klyshko equivalent optical path. Subsequently, from the information-theoretic perspective, the ghost imaging process is modeled as the transmission of information in a “source–channel–sink” model: the applicable conditions and internal relations of classical information metrics, including mutual information, channel capacity, Fisher information, and the Cramér–Rao bound, in imaging performance evaluation are analyzed; the encoding optimization and decoding optimization approaches guided by information metrics are classified, compared, and reviewed; and the applicable conditions and feasible paths for introducing quantum information-theoretic tools such as quantum mutual information and quantum Fisher information are explored. It is pointed out that the statistical structure of the second-order correlation function of the optical field determines the information acquisition capability of the imaging system, and that information metrics are evolving from “post-hoc evaluation” toward “a priori design”. It is concluded that mutual information, channel capacity, and Fisher information characterize imaging performance in terms of information acquisition efficiency, channel transmission potential, and parameter estimation precision, respectively, and are mutually complementary. It is proposed that future research be deepened in directions such as the collaborative design of encoding and decoding optimization, the quantitative comparison between the performance of decoding algorithms and information-theoretic limits, quantum channel modeling for ghost imaging with non-classical light sources, and applications oriented toward weak signals and special bands, so as to further reveal the mechanisms of information transfer in the ghost imaging process, enhance the information transmission efficiency and image reconstruction accuracy of imaging systems, and promote the application and development of related technologies in biomedical imaging, radar detection, and other fields.