Abstract: Optical correlated recognition is one of the important methods in image recognition applications. For optical correlator, to effectively recognize the peak signal of correlated output plane is the key factor to ensure the accurate image recognition. The traditional threshold method can’t achieve satisfactory results due to the output power fluctuation of lasers, errors from optical systems and noise inherent in SLMs. The author proposed that, in order to effectively classify and recognize the correlated peak signal and noise to improve the performance of the optical system, the output plane of the correlator should be preprocessed, the shape information of correlating signal should be well considered, the ROI (range of interest) should be extracted and the BP neural network should be adopted to calculate the input vector. The result shows that the proposed method can improve the reliability of the correlator and reduce the possibility of misjudgments.