Can N-dimensional convolutional neural networks distinguish men and women better than humans do?

Convolutional neural networks (CNN) were able to beat human performance in various areas of 2D image recognition, e.g., in the German Traffic Sign competition run by IJCNN 2011. While the majority of classical image processing techniques is based on carefully pre-selected image features, CNNs are designed to learn local features autonomously. A growing availability of high-dimensional object data, e.g., from medicine or forensic analysis, thus motivated us to develop a new variant of the classical CNN model. The introduced N-dimensional convolutional neural networks (ND-CNN) enhanced with an enforced internal knowledge representation allow to process general N-dimensional object data while supporting adequate interpretation of the found object characteristics. Experimental results obtained so far for gender classification of 3D face scans confirm an extremely strong power of the proposed neural classifier. The developed ND-CNNs significantly outperformed humans (by 33%) while still allowing for a transparent representation of the face features present and detected in the data.