Manuel Günther

Bob: a free signal processing and machine learning toolbox for researchers

Bob is a free signal processing and machine learning toolbox originally developed by the Biometrics group at Idiap Research Institute, Switzerland. The toolbox is designed to meet the needs of researchers by reducing development time and efficiently processing data. Firstly, Bob provides a researcher-friendly Python environment for rapid development. Secondly, efficient processing of large amounts of multimedia data is provided by fast C++ implementations of identified bottlenecks. The Python environment is integrated seamlessly with the C++ library, which ensures the library is easy to use and extensible. Thirdly, Bob supports reproducible research through its integrated experimental protocols for several databases. Finally, a strong emphasis is placed on code clarity, documentation, and thorough unit testing. Bob is thus an attractive resource for researchers due to this unique combination of ease of use, efficiency, extensibility and transparency. Bob is an open-source library and an ongoing community effort.

Genetic determination of human facial morphology: links between cleft-lips and normal variation

Recent genome-wide association studies have identified single nucleotide polymorphisms (SNPs) associated with non-syndromic cleft lip with or without cleft palate (NSCL/P), and other previous studies showed distinctly differing facial distance measurements when comparing unaffected relatives of NSCL/P patients with normal controls. Here, we test the hypothesis that genetic loci involved in NSCL/P also influence normal variation in facial morphology. We tested 11 SNPs from 10 genomic regions previously showing replicated evidence of association with NSCL/P for association with normal variation of nose width and bizygomatic distance in two cohorts from Germany (N=529) and the Netherlands (N=2497). The two most significant associations found were between nose width and SNP rs1258763 near the GREM1 gene in the German cohort (P=6 × 10−4), and between bizygomatic distance and SNP rs987525 at 8q24.21 near the CCDC26 gene (P=0.017) in the Dutch sample. A genetic prediction model explained 2% of phenotype variation in nose width in the German and 0.5% of bizygomatic distance variation in the Dutch cohort. Although preliminary, our data provide a first link between genetic loci involved in a pathological facial trait such as NSCL/P and variation of normal facial morphology. Moreover, we present a first approach for understanding the genetic basis of human facial appearance, a highly intriguing trait with implications on clinical practice, clinical genetics, forensic intelligence, social interactions and personal identity.

A novel approach to the detection of acromegaly: accuracy of diagnosis by automatic face classification.

CONTEXT The delay between onset of first symptoms and diagnosis of the acromegaly is 6-10 yr. Acromegaly causes typical changes of the face that might be recognized by face classification software. OBJECTIVE The objective of the study was to assess classification accuracy of acromegaly by face-classification software. DESIGN This was a diagnostic study. SETTING The study was conducted in specialized care. PARTICIPANTS Participants in the study included 57 patients with acromegaly (29 women, 28 men) and 60 sex- and age-matched controls. INTERVENTIONS We took frontal and side photographs of the faces and grouped patients into subjects with mild, moderate, and severe facial features of acromegaly by overall impression. We then analyzed all pictures using computerized similarity analysis based on Gabor jets and geometry functions. We used the leave-one-out cross-validation method to classify subjects by the software. Additionally, all subjects were classified by visual impression by three acromegaly experts and three general internists. MAIN OUTCOME MEASURE Classification accuracy by software, experts, and internists was measured. FINDINGS The software correctly classified 71.9% of patients and 91.5% of controls. Classification accuracy for patients by visual analysis was 63.2 and 42.1% by experts and general internists, respectively. Classification accuracy for controls was 80.8 and 87.0% by experts and internists, respectively. The highest differences in accuracy between software and experts and internists were present for patients with mild acromegaly. CONCLUSIONS Acromegaly can be detected by computer software using photographs of the face. Classification accuracy by software is higher than by medical experts or general internists, particularly in patients with mild features of acromegaly. This is a promising tool to help detecting acromegaly.

The 2013 speaker recognition evaluation in mobile environment

This paper evaluates the performance of the twelve primary systems submitted to the evaluation on speaker verification in the context of a mobile environment using the MOBIO database. The mobile environment provides a challenging and realistic test-bed for current state-of-the-art speaker verification techniques. Results in terms of equal error rate (EER), half total error rate (HTER) and detection error trade-off (DET) confirm that the best performing systems are based on total variability modeling, and are the fusion of several sub-systems. Nevertheless, the good old UBM-GMM based systems are still competitive. The results also show that the use of additional data for training as well as gender-dependent features can be helpful.

MOON: A Mixed Objective Optimization Network for the Recognition of Facial Attributes

Attribute recognition, particularly facial, extracts many labels for each image. While some multi-task vision problems can be decomposed into separate tasks and stages, e.g., training independent models for each task, for a growing set of problems joint optimization across all tasks has been shown to improve performance. We show that for deep convolutional neural network (DCNN) facial attribute extraction, multi-task optimization is better. Unfortunately, it can be difficult to apply joint optimization to DCNNs when training data is imbalanced, and re-balancing multi-label data directly is structurally infeasible, since adding/removing data to balance one label will change the sampling of the other labels. This paper addresses the multi-label imbalance problem by introducing a novel mixed objective optimization network (MOON) with a loss function that mixes multiple task objectives with domain adaptive re-weighting of propagated loss. Experiments demonstrate that not only does MOON advance the state of the art in facial attribute recognition, but it also outperforms independently trained DCNNs using the same data. When using facial attributes for the LFW face recognition task, we show that our balanced (domain adapted) network outperforms the unbalanced trained network.

Bi-modal biometric authentication on mobile phones in challenging conditions

This paper examines the issue of face, speaker and bi-modal authentication in mobile environments when there is significant condition mismatch. We introduce this mismatch by enrolling client models on high quality biometric samples obtained on a laptop computer and authenticating them on lower quality biometric samples acquired with a mobile phone. To perform these experiments we develop three novel authentication protocols for the large publicly available MOBIO database. We evaluate state-of-the-art face, speaker and bi-modal authentication techniques and show that inter-session variability modelling using Gaussian mixture models provides a consistently robust system for face, speaker and bi-modal authentication. It is also shown that multi-algorithm fusion provides a consistent performance improvement for face, speaker and bi-modal authentication. Using this bi-modal multi-algorithm system we derive a state-of-the-art authentication system that obtains a half total error rate of 6.3% and 1.9% for Female and Male trials, respectively.

An open source framework for standardized comparisons of face recognition algorithms

In this paper we introduce the facereclib, the first software library that allows to compare a variety of face recognition algorithms on most of the known facial image databases and that permits rapid prototyping of novel ideas and testing of meta-parameters of face recognition algorithms. The facereclib is built on the open source signal processing and machine learning library Bob. It uses well-specified face recognition protocols to ensure that results are comparable and reproducible. We show that the face recognition algorithms implemented in Bob as well as third party face recognition libraries can be used to run face recognition experiments within the framework of the facereclib. As a proof of concept, we execute four different state-of-the-art face recognition algorithms: local Gabor binary pattern histogram sequences (LGBPHS), Gabor graph comparisons with a Gabor phase based similarity measure, inter-session variability modeling (ISV) of DCT block features, and the linear discriminant analysis on two different color channels (LDA-IR) on two different databases: The Good, The Bad, and The Ugly, and the BANCA database, in all cases using their fixed protocols. The results show that there is not one face recognition algorithm that outperforms all others, but rather that the results are strongly dependent on the employed database.

Face recognition with disparity corrected gabor phase differences

We analyze the relative relevance of Gabor amplitudes and phases for face recognition. We propose an algorithm to reliably estimate offset point disparities from phase differences and show that disparity-corrected Gabor phase differences are well suited for face recognition in difficult lighting conditions. The method reaches 74.8% recognition rate on the Lighting set of the CAS-PEAL database and 35.7% verification rate on experiment 2.4 of the FRGC database.

A Survey of Stealth Malware: Attacks, Mitigation Measures, and Steps Toward Autonomous Open World Solutions

As our professional, social, and financial existences become increasingly digitized and as our government, healthcare, and military infrastructures rely more on computer technologies, they present larger and more lucrative targets for malware. Stealth malware in particular poses an increased threat because it is specifically designed to evade detection mechanisms, spreading dormant, in the wild for extended periods of time, gathering sensitive information or positioning itself for a high-impact zero-day attack. Policing the growing attack surface requires the development of efficient anti-malware solutions with improved generalization to detect novel types of malware and resolve these occurrences with as little burden on human experts as possible. In this paper, we survey malicious stealth technologies as well as existing solutions for detecting and categorizing these countermeasures autonomously. While machine learning offers promising potential for increasingly autonomous solutions with improved generalization to new malware types, both at the network level and at the host level, our findings suggest that several flawed assumptions inherent to most recognition algorithms prevent a direct mapping between the stealth malware recognition problem and a machine learning solution. The most notable of these flawed assumptions is the closed world assumption: that no sample belonging to a class outside of a static training set will appear at query time. We present a formalized adaptive open world framework for stealth malware recognition and relate it mathematically to research from other machine learning domains.

Score Calibration in Face Recognition

An evaluation of the verification and calibration performance of a face recognition system based on inter-session variability modelling is presented. As an extension to calibration through linear transformation of scores, categorical calibration is introduced as a way to include additional information about images for calibration. The cost of likelihood ratio, which is a well-known measure in the speaker recognition field, is used as a calibration performance metric. The results obtained from the challenging mobile biometrics and surveillance camera face databases indicate that linearly calibrated face recognition scores are less misleading in their likelihood ratio interpretation than uncalibrated scores. In addition, the categorical calibration experiments show that calibration can be used not only to improve the likelihood ratio interpretation of scores, but also to improve the verification performance of a face recognition system.