The work aims to propose a fault diagnosis and locating method based on image analysis of sound field with respect to the difficulty in acquiring and diagnosing fault information of ink roller bearing in printing press. Firstly, the sound test system was constructed and the space area of detected object was divided. The signals were acquired one by one for reconstruction and combination according to the corresponding spatial location relationship. Then, the signal energy in a period of time was superimposed and the sound field image information in the superposition state was formed based on the spatial location relationship. Next, bidimensional empirical mode decomposition (BEMD) was introduced to analyze the spatial domain image information and get the high-frequency information through the elimination of ambient noise interference. Feature analysis and comparison of fault bearings for the ink roller system at different positions and of different kinds were carried out. The diagnosis of faults in inner and outer rings of the bearing was realized by means of IBIMF component and statistical characteristics. The effective locating of different fault positions was achieved through dimensionless characteristics and the effectiveness of the method was verified by experiments. The proposed fault diagnosis method makes the most of non-contact measurement and has a strong ability to characterize the faults, which is suitable for similar complex mechanical systems.