Aiming at the problem that the modal aliasing phenomenon of Local Mean Decomposition (LMD) method in the decomposition process of nonlinear and non-stationary vibration signals, affects the accuracy of identification, a fault diagnosis method based on Conditional Local Mean Decomposition (CLMD) method and Variable Predictive Model Class Discriminate (VPMCD) was proposed. The method combined the frequency resolution method of digital image processing with LMD. Firstly, the frequency resolutions of all local extreme points were calculated, and according to the frequency resolutions of local extreme points, the vibration signals could be divided into the low frequency resolution area and the high frequency resolution area. Secondly, LMD method was used to decompose the high frequency resolution area to get several components of Product Function (PF). Finally, after these PF components were connected by broken line, PF could be got through moving average processing. The skewness coefficient and the energy coefficient of PF could form fault feature vector. VPMCD could use fault feature vector to identify the fault types. This method was applied into bearing fault diagnosis. The experimental results show that the recognition efficiency of the proposed method increases by 8.33%, compared with LMD. As a result, the method is feasible and valid.