Tim Oblak

The large majority of modern software solutions intended for fingermark processing in a forensic context is heavily dependant on the correct image scaling. Fingermark images captured with digital cameras at a crime scene require the use of physical rulers or labels. While the resolution of a fingermark image can be calibrated manually by a forensic examiner in a lab, we propose an automated approach, which could be integrated directly into existing identification systems and would eliminate the need for human intervention. Our approach consists of a CNN regressor, which directly predicts the PPI of stochastically-sampled local patches based on the friction ridge information contained within. In a range of PPI between 500 and 1500, our method achieves a mean average error of around 24 PPI for fingerprint and fingermark images.

The assessment of fingermark (latent fingerprint) quality is an intrinsic part of a forensic investigation. The fingermark quality indicates the value and utility of the trace evidence recovered from the crime scene in the course of a forensic investigation; it determines how the evidence will be processed, and it correlates with the probability of finding a corresponding fingerprint in the reference dataset. The deposition of fingermarks on random surfaces occurs spontaneously in an uncontrolled fashion, which introduces imperfections to the resulting impression of the friction ridge pattern. In this work, we propose a new probabilistic framework for Automated Fingermark Quality Assessment (AFQA). We used modern deep learning techniques, which have the ability to extract patterns even from noisy data, and combined them with a methodology from the field of eXplainable AI (XAI) to make our models more transparent. Our solution first predicts a quality probability distribution, from which we then calculate the final quality value and, if needed, the uncertainty of the model. Additionally, we complemented the predicted quality value with a corresponding quality map. We used GradCAM to determine which regions of the fingermark had the largest effect on the overall quality prediction. We show that the resulting quality maps are highly correlated with the density of minutiae points in the input image. Our deep learning approach achieved high regression performance, while significantly improving the interpretability and transparency of the predictions.

Reconstruction of 3D space from visual data has always been a significant challenge in the field of computer vision. A popular approach to address this problem can be found in the form of bottom-up reconstruction techniques which try to model complex 3D scenes through a constellation of volumetric primitives. Such techniques are inspired by the current understanding of the human visual system and are, therefore, strongly related to the way humans process visual information, as suggested by recent visual neuroscience literature. While advances have been made in recent years in the area of 3D reconstruction, the problem remains challenging due to the many possible ways of representing 3D data, the ambiguity of determining the shape and general position in 3D space and the difficulty to train efficient models for the prediction of volumetric primitives. In this paper, we address these challenges and present a novel solution for recovering volumetric primitives from depth images. Specifically, we focus on the recovery of superquadrics, a special type of parametric models able to describe a wide array of 3D shapes using only a few parameters. We present a new learning objective that relies on the superquadric (inside-outside) function and develop two learning strategies for training convolutional neural networks (CNN) capable of predicting superquadric parameters. The first uses explicit supervision and penalizes the difference between the predicted and reference superquadric parameters. The second strategy uses implicit supervision and penalizes differences between the input depth images and depth images rendered from the predicted parameters. CNN predictors for superquadric parameters are trained with both strategies and evaluated on a large dataset of synthetic and real-world depth images. Experimental results show that both strategies compare favourably to the existing state-of-the-art and result in high quality 3D reconstructions of the modelled scenes at a much shorter processing time.