Research on Image Super-Resolution Reconstruction Based on BPNN and RBFNN

Abstract

To overcome inherent resolution limitation of imaging system and provide a high resolution satellite image economically and practically, the paper researched on image super-resolution reconstruction (SRR) based on BPNN and RBFNN. Learning sample images are obtained according to image observation model, input/output mapping pairs are formed by way of vector mappings, and then two NN models of SRR are established respectively. Simulation and generalization tests are carried on the BPNN and RBFNN respectively. Results indicate that reconstruction performance of BPNN is better than that of RBFNN in aspect of reconstruction time-consuming, convergence speed and effect of SRR in generalization tests, although those parameters are better in simulation tests of RBFNN. So BPNN is more suitable for project applications of image super-resolution reconstruction.