The traditional superresolution algorithm based on convolutional network is difficult to recover highresolution images and fuse edge information in different scenes. In order to solve this problem and based on the detailed analysis of the network of the typical models, the parallel mapping convolutional network proposed model relies on the analysis of the problem of reconstruction module inputs and loss function constraints. The model based on endtoend manner, constructing parallel mapping convolutional network and regularization constraints, can be hierarchically independent of image features extracted and greatly enrich image in polymerization of high resolution image feature dimension. Meanwhile, the network introduces the total variation regularization after the convolution layer and constraints illposed problem, which extract accurate and robust image features from the network, and enrich the edge information of image. The experimental results on typical databases show that the proposed algorithm achieves better superresolution results, the subjective visual effect and objective evaluation indices are improved significantly, and the resolution of the image is enhanced.