Temporal Image Forensics: Using CNNs for a Chronological Ordering of Line-Scan Data

We propose to use CNNs for obtaining a temporal ordering of line-scanner data. Excellent age classification accuracy is achieved only in case network training and testing is done with image patches taken from consistent spatial locations, i.e. temporal features exploited are bound to specific positions in the image. With spatially consistent patches, up to 100% classification accuracy can be achieved, whereas with spatially varying patches the accuracy stagnates at around 54%. We have also noted a result dependency on image content and have found, that a consistent patch position relative to the scanning line is not sufficient for good results.