In order to get good general performance of optical joint transform correlation pattern recognition, a new reference shifted phase-encoded fringe-adjusted joint transform correlator for distortion invariant pattern recognition is proposed based on Sliced Orthogonal Nonlinear Generalized (SONG) correlation. The new correlator not only enhances the performance of resisting disturbance signal and noise, but also has a good performance of multiple target recognition by introducing the technique of SONG correlation and Synthetic Discriminant Function (SDF). It realizes distortion invariant pattern recognition by modifying the constraint condition of SDF when several distorted targets occur in the input image. The results of simulation indicate that correlation peak value and peak-to-noise ratio of output is 1.87~5.16 times and 1.88~5.19 times of those without introduction of SONG correlation respectively during whole noise density intervals when input image is destroyed by nonoverlapping background noise. And average deviation is less than 0.04. Preferable correlation peak value and peak-to-noise ratio of output can be obtained during most of noise density intervals when input image is destroyed by substitutive noise, and average deviation is less than 0.1 when noise density is not more than 0.4.