Based on different projection geometry, a fisheye image can be presented as a parameterized non-rectilinear image. Deep neural networks(DNN) is one of the solutions for extracting the parameters for fisheye image feature expression. However, a number of images are required for training a reasonable prediction model for DNN. In this paper, we propose to extend the scale of the dataset using parameterized synthetic images which effectively boost the diversity of samples and avoid the limitation on the scale. To simulate different viewing angles and distances, we adopt controllable parameterized projection processes on transformation. The reliability of the proposed method is tested with the image captured by a fisheye camera. The synthetic fisheye image dataset is the first dataset that is developed by existing labeled perspective images. It is accessible via: http://www2.leuphana.de/misl/fisheye-data-set/.