Automatic selection of Current Image (CI) is a key technology for automation of scene matching navigation system. An algorithm for automatic selection of CI based on 3-layers BP net integrated with Sobel template and on neural network was proposed. The simulation results indicated that the algorithm is simple in structure, short in time, and highly accurate in image classification. By analysis of testing samples, BP parallel structure was shown to be achievable in real time implementation. As tested by Least Square and Bayes Scene Matching algorithms, the current image could reach a final precision higher than 0.5 pixel by means of BP net automatic selection algorithm, simple Scene Matching and accurate Scene Matching.