This paper targets a new support process for inspiration in civil engineering design. First the notion of inspiration within the engineering design process is briefly explained, and existing approaches for automated creativity support are reviewed. Further the definition and procedural layout of the “inspiration mining” process is introduced, and hypotheses on “inspiration objects” and key processes (matching of inspiration objects and engineering tasks) are formulated. The semantic and visual distance of these objects to the engineering challenge is a crucial factor and needs to be quantified to retrieve relevant concepts. In order to map visual similarity within a sample from the WikiArt dataset, we train a Convolutional Neural Network in an autoencoder setup on the data. The generated image embedding is used to compare it to the approach of Transfer Learning on the same dataset by a pre-trained neural network model (VGG19 on ImageNet dataset).