Improving the performance of multimedia pedestrian classification with images synthesized using a deep convolutional generative adversarial network