Face targets in image taken by the current surveillance video are small and difficult to identify. The image super resolution processing has become the technology and means to solve the image practical application problems of surveillance video. A super resolution technology for outdoor surveillance video face image is proposed. The prior knowledge is used to construct the image training set, and some pre-processing operations like the image space conversion and denoising are operated. The convolutional neural network with eight layers is designed, and its layer types and connection mode are set. Meanwhile, the activation function types and the transmission mode functions among layers are set. The network parameters are initialized and the network is trained according to the training set. The convolution kernels and bias parameters are adjusted reversely by the loss function, and the image output is implemented. Through a large number of actual monitoring video image tests, and compared with other existing methods, the experimental results show that the proposed method has certain advantages in the effect of image super resolution and processing speed.