Incremental learning of feature space and classifier for face recognition.

We have proposed a new approach to pattern recognition in which not only a classifier but also a feature space of input variables is learned incrementally. In this paper, an extended version of Incremental Principal Component Analysis (IPCA) and Resource Allocating Network with Long-Term Memory (RAN-LTM) are effectively combined to implement this idea. Since IPCA updates a feature space incrementally by rotating the eigen-axes and increasing the dimensions, the inputs of a neural classifier must also change in their values and the number of input variables. To solve this problem, we derive an approximation of the update formula for memory items, which correspond to representative training samples stored in the long-term memory of RAN-LTM. With these memory items, RAN-LTM is efficiently reconstructed and retrained to adapt to the evolution of the feature space. This function is incorporated into our face recognition system. In the experiments, the proposed incremental learning model is evaluated over a self-compiled video clip of 24 persons. The experimental results show that the incremental learning of a feature space is very effective to enhance the generalization performance of a neural classifier in a realistic face recognition task.     

Hybrid extreme rotation forest

This paper proposes the Hybrid Extreme Rotation Forest (HERF), an innovative ensemble learning algorithm for classification problems, combining classical Decision Trees with the recently proposed Extreme Learning Machines (ELM) training of Neural Networks. In the HERF algorithm, training of each individual classifier involves two steps: first computing a randomized data rotation transformation of the training data, second, training the individual classifier on the rotated data. The testing data is subjected to the same transformation as the training data, which is specific for each classifier in the ensemble. Experimental design in this paper involves (a) the comparison of factorization approaches to compute the randomized rotation matrix: the Principal Component Analysis (PCA) and the Quartimax, (b) assessing the effect of data normalization and bootstrapping training data selection, (c) all variants of single and combined ELM and decision trees, including Regularized ELM. This experimental design effectively includes other state-of-the-art ensemble approaches in the comparison, such as Voting ELM and Random Forest. We report extensive results over a collection of machine learning benchmark databases. Ranking the cross-validation results per experimental dataset and classifier tested concludes that HERF significantly improves over the other state-of-the-art ensemble classifier. Besides, we find some other results such as that the data rotation with Quartimax improves over PCA, and the relative insensitivity of the approach to regularization which may be attributable to the de facto regularization performed by the ensemble approach.     

Enhancement of face recognition learning in patients with brain injury using three cognitive training procedures

Twenty patients with impairments of face recognition, in the context of a broader pattern of cognitive deficits, were administered three new training procedures derived from contemporary theories of face processing to enhance their learning of new faces: semantic association (being given additional verbal information about the to-be-learned faces); caricaturing (presentation of caricatured versions of the faces during training and veridical versions at recognition testing); and part recognition (focusing patients on distinctive features during the training phase). Using a within-subjects design, each training procedure was applied to a different set of 10 previously unfamiliar faces and entailed six presentations of each face. In a “simple exposure” control procedure (SE), participants were given six presentations of another set of faces using the same basic protocol but with no further elaboration. Order of the four procedures was counterbalanced, and each condition was administered on a different day. A control group of 12 patients with similar levels of face recognition impairment were trained on all four sets of faces under SE conditions.

Compared to the SE condition, all three training procedures resulted in more accurate discrimination between the 10 studied faces and 10 distractor faces in a post-training recognition test. This did not reflect any intrinsic lesser memorability of the faces used in the SE condition, as evidenced by the comparable performance across face sets by the control group.

At the group level, the three experimental procedures were of similar efficacy, and associated cognitive deficits did not predict which technique would be most beneficial to individual patients; however, there was limited power to detect such associations. Interestingly, a pure prosopagnosic patient who was tested separately showed benefit only from the part recognition technique. Possible mechanisms for the observed effects, and implications for rehabilitation, are discussed.     

Enhancement of face recognition learning in patients with brain injury using three cognitive training procedures

Twenty patients with impairments of face recognition, in the context of a broader pattern of cognitive deficits, were administered three new training procedures derived from contemporary theories of face processing to enhance their learning of new faces: semantic association (being given additional verbal information about the to-be-learned faces); caricaturing (presentation of caricatured versions of the faces during training and veridical versions at recognition testing); and part recognition (focusing patients on distinctive features during the training phase). Using a within-subjects design, each training procedure was applied to a different set of 10 previously unfamiliar faces and entailed six presentations of each face. In a "simple exposure" control procedure (SE), participants were given six presentations of another set of faces using the same basic protocol but with no further elaboration. Order of the four procedures was counterbalanced, and each condition was administered on a different day. A control group of 12 patients with similar levels of face recognition impairment were trained on all four sets of faces under SE conditions. Compared to the SE condition, all three training procedures resulted in more accurate discrimination between the 10 studied faces and 10 distractor faces in a post-training recognition test. This did not reflect any intrinsic lesser memorability of the faces used in the SE condition, as evidenced by the comparable performance across face sets by the control group. At the group level, the three experimental procedures were of similar efficacy, and associated cognitive deficits did not predict which technique would be most beneficial to individual patients; however, there was limited power to detect such associations. Interestingly, a pure prosopagnosic patient who was tested separately showed benefit only from the part recognition technique. Possible mechanisms for the observed effects, and implications for rehabilitation, are discussed.     

Short utterance recognition using a network with minimum training

A feedforward network is used to recognize short, digitized, isolated utterances. A high, multispeaker recognition rate is achieved with a small vocabulary with a single training utterance. This approach makes use of the pattern recognition property of the network architecture to classify different temporal patterns in the multidimensional feature space. The network recognizes the utterances without the need of segmentation, phoneme identification, or time alignment. We train the network with four words spoken by one single speaker. The network is then able to recognize 20 tokens spoken by 5 other speakers. We repeat the above training and testing procedure using a different speaker's utterances for training each time. The overall accuracy is 97.5%. We compare this approach to the traditional dynamic programming (DP) approach, and find that DP with slope constraints of 0 and 1 achieve 98.5% and 85% accuracies respectively. Finally we validate out statistics by training and testing the network of a four-word subset of the Texas Instruments (Tl) isolated word database. The accuracy with this vocabulary exceeds 96%. By doubling the size of the training set, the accuracy is raised to 98%. Using a suitable threshold, we are able to raise the accuracy of one network from 87% to 98.5%. Thresholding applied to all networks would then raise the overall accuracy to well over 99%.

This technique is especially promising because of the low overhead and computational requirements, which make it suitable for a low cost, portable, command recognition type of application.     

Does acquisition of Greeble expertise in prosopagnosia rule out a domain-general deficit?

According to the expertise account of face specialization, a deficit that affects general expertise mechanisms should similarly impair the expert individuation of both faces and other visually homogeneous object classes. To test this possibility, we attempted to train a prosopagnosic patient, LR, to become a Greeble expert using the standard Greeble expertise-training paradigm (Gauthier & Tarr, 2002). Previous research demonstrated that LR's prosopagnosia was related to an inability to simultaneously use multiple features in a speeded face recognition task (Bukach, Bub, Gauthier, & Tarr, 2006). We hypothesized that LR's inability to use multiple face features would manifest in his acquisition of Greeble expertise, even though his basic object recognition is unimpaired according to standard neuropsychological testing. Although LR was eventually able to reach expertise criterion, he took many more training sessions than controls, suggesting use of an abnormal strategy. To further explore LR's Greeble processing strategies, we assessed his ability to use multiple Greeble features both before and after Greeble training. LR's performance in two versions of this task demonstrates that, even after training, he relies heavily on a single feature to identify Greebles. This correspondence between LR's face recognition and post-training Greeble recognition supports the idea that impaired face recognition is simply the most visible symptom of a more general object recognition impairment in acquired prosopagnosia.     

A New Hand Image Database Simultaneously Acquired in Visible,Near-Infrared and Thermal Spectrums

In this paper, we present a new hand database called Tecnocampus Hand Image Database that includes right hand, palm and dorsal images. All the images have been acquired with three different sensors (visible, near-infrared and thermal). This database consists of 100 people acquired in five different acquisition sessions, two images per session and palm/dorsal sides. The total amount of pictures is 6.000, and it is mainly developed for hand image biometric recognition purposes. In addition, the database has been studied from the information theory point of view, and we found that this highest level of information is achieved in thermal spectrum. Furthermore, a low level of mutual information between different spectrums is also demonstrated. This opens an interesting research field in multi-sensor data fusion.     

Online decoding of object‐based attention using real‐time fMRI

Visual attention is used to selectively filter relevant information depending on current task demands and goals. Visual attention is called object‐based attention when it is directed to coherent forms or objects in the visual field. This study used real‐time functional magnetic resonance imaging for moment‐to‐moment decoding of attention to spatially overlapped objects belonging to two different object categories. First, a whole‐brain classifier was trained on pictures of faces and places. Subjects then saw transparently overlapped pictures of a face and a place, and attended to only one of them while ignoring the other. The category of the attended object, face or place, was decoded on a scan‐by‐scan basis using the previously trained decoder. The decoder performed at 77.6% accuracy indicating that despite competing bottom‐up sensory input, object‐based visual attention biased neural patterns towards that of the attended object. Furthermore, a comparison between different classification approaches indicated that the representation of faces and places is distributed rather than focal. This implies that real‐time decoding of object‐based attention requires a multivariate decoding approach that can detect these distributed patterns of cortical activity.     

The Effect of Fatigue on the Performance of Online Writer Recognition

Cognitive Computation - The performance of biometric modalities based on things done by the subject, like signature and text-based recognition, may be affected by the subject’s state. Fatigue...     

Fat neural network for recognition of position-normalised objects

The design of a recognition system for natural objects is difficult, mainly because such objects are subject to a strong variability that cannot be easily modelled: planktonic species possess such highly variable forms. Existing plankton recognition systems usually comprise feature extraction processing upstream of a classifier. Drawbacks of such an approach are that the design of relevant feature extraction processes may be very difficult, especially if classes are numerous and if intra-class variability is high, so that the system becomes specific to the problem for which features have been tuned. The opposite course that we take is based on a structured multi-layer neural network with no shared weights, which generates its own features during training. Such a large parameterised—fat—network exhibits good generalisation capabilities for pattern recognition problems dealing with position-normalised objects, even with as many as one thousand weights as training examples. The advantage of such large networks, in terms of generalisation efficiency, adaptability and classification time, is demonstrated by applying the network to three plankton recognition and face recognition problems. Its ability to perform good generalisation with few training examples, but many weights, is an open theoretical problem.     

Fast Image Recognition Based on Independent Component Analysis and Extreme Learning Machine

Nowadays, image recognition has become a highly active research topic in cognitive computation community, due to its many potential applications. Generally, the image recognition task involves two subtasks: image representation and image classification. Most feature extraction approaches for image representation developed so far regard independent component analysis (ICA) as one of the essential means. However, ICA has been hampered by its extremely expensive computational cost in real-time implementation. To address this problem, a fast cognitive computational scheme for image recognition is presented in this paper, which combines ICA and the extreme learning machine (ELM) algorithm. It tries to solve the image recognition problem at a much faster speed by using ELM not only in image classification but also in feature extraction for image representation. As an example, our proposed approach is applied to the face image recognition with detailed analysis. Firstly, common feature hypothesis is introduced to extract the common visual features from universal images by the traditional ICA model in the offline recognition process, and then ELM is used to simulate ICA for the purpose of facial feature extraction in the online recognition process. Lastly, the resulting independent feature representation of the face images extracted by ELM rather than ICA will be fed into the ELM classifier, which is composed of numerous single hidden layer feed-forward networks. Experimental results on Yale face database and MNIST digit database have shown the good performance of our proposed approach, which could be comparable to the state-of-the-art techniques at a much faster speed.     

A Comparative Study of In-Air Trajectories at Short and Long Distances in Online Handwriting

Existing literature about online handwriting analysis to support pathology diagnosis has taken advantage of in-air trajectories. A similar situation occurred in biometric security applications where the goal is to identify or verify an individual using his signature or handwriting. These studies do not consider the distance of the pen tip to the writing surface. This is due to the fact that current acquisition devices do not provide height formation. However, it is quite straightforward to differentiate movements at two different heights (a) short distance: height lower or equal to 1 cm above a surface of digitizer, the digitizer provides x and y coordinates; (b) long distance: height exceeding 1 cm, the only information available is a time stamp that indicates the time that a specific stroke has spent at long distance. Although short distance has been used in several papers, long distances have been ignored and will be investigated in this paper. In this paper, we will analyze a large set of databases (BIOSECUR-ID, EMOTHAW, PaHaW, OXYGEN-THERAPY, and SALT), which contain a total amount of 663 users and 17,951 files. We have specifically studied (a) the percentage of time spent on-surface, in-air at short distance, and in-air at long distance for different user profiles (pathological and healthy users) and different tasks; (b) the potential use of these signals to improve classification rates. Our experimental results reveal that long distance movements represent a very small portion of the total execution time (0.5% in the case of signatures and 10.4% for uppercase words of BIOSECUR-ID, which is the largest database). In addition, significant differences have been found in the comparison of pathological versus control group for letter “l” in PaHaW database (p = 0.0157) and crossed pentagons in SALT database (p = 0.0122).     

Pipelining of Fuzzy ARTMAP without matchtracking: correctness, performance bound, and Beowulf evaluation.

Fuzzy ARTMAP neural networks have been proven to be good classifiers on a variety of classification problems. However, the time that Fuzzy ARTMAP takes to converge to a solution increases rapidly as the number of patterns used for training is increased. In this paper we examine the time Fuzzy ARTMAP takes to converge to a solution and we propose a coarse grain parallelization technique, based on a pipeline approach, to speed-up the training process. In particular, we have parallelized Fuzzy ARTMAP without the match-tracking mechanism. We provide a series of theorems and associated proofs that show the characteristics of Fuzzy ARTMAP's, without matchtracking, parallel implementation. Results run on a BEOWULF cluster with three large databases show linear speedup as a function of the number of processors used in the pipeline. The databases used for our experiments are the Forrest CoverType database from the UCI Machine Learning repository and two artificial databases, where the data generated were 16-dimensional Gaussian distributed data belonging to two distinct classes, with different amounts of overlap (5% and 15%).     

Dissociation between face perception and face memory in adults,but not children,with developmental prosopagnosia

Cognitive models propose that face recognition is accomplished through a series of discrete stages, including perceptual representation of facial structure, and encoding and retrieval of facial information. This implies that impaired face recognition can result from failures of face perception, face memory, or both. Studies of acquired prosopagnosia, autism spectrum disorders, and the development of normal face recognition support the idea that face perception and face memory are distinct processes, yet this distinction has received little attention in developmental prosopagnosia (DP). To address this issue, we tested the face perception and face memory of children and adults with DP. By definition, face memory is impaired in DP, so memory deficits were present in all participants. However, we found that all children, but only half of the adults had impaired face perception. Thus, results from adults indicate that face perception and face memory are dissociable, while the results from children provide no evidence for this division. Importantly, our findings raise the possibility that DP is qualitatively different in childhood versus adulthood. We discuss theoretical explanations for this developmental pattern and conclude that longitudinal studies are necessary to better understand the developmental trajectory of face perception and face memory deficits in DP.     

Deep Transfer Learning on the Aggregated Dataset for Face Presentation Attack Detection

Presentation attacks are becoming a serious threat to one of the most common biometric applications, namely face recognition (FR). In recent years, numerous methods have been presented to detect and identify these attacks using publicly available datasets. However, such datasets are often collected in controlled environments and are focused on one specific type of attack. We hypothesise that a model’s accurate performance on one or more public datasets does not necessarily guarantee generalisation across other, unseen face presentation attacks. To verify our hypothesis, in this paper, we present an experimental framework where the generalisation ability of pre-trained deep models is assessed using four popular and commonly used public datasets. Extensive experiments were carried out using various combinations of these datasets. Results show that, in some circumstances, a slight improvement in model performance can be achieved by combining different datasets for training purposes. However, even with a combination of public datasets, models still could not be trained to generalise to unseen attacks. Moreover, models could not necessarily generalise to a learned format of attack over different datasets. The work and results presented in this paper suggest that more diverse datasets are needed to drive this research as well as the need for devising new methods capable of extracting spoof-specific features which are independent of specific datasets.

     

An Information Analysis of In-Air and On-Surface Trajectories in Online Handwriting

This paper is aimed at analysing, from an information theory perspective, the gestures produced by human beings when handwriting a text. Modern capturing devices allow the gathering of data not only from the on-surface movements of the hand, but also from the in-air trajectories performed when the hand moves in the air from one stroke to the next. Our past research with isolated uppercase words clearly suggests that both types of trajectories have a biometric potential to perform writer recognition and that they can be effectively combined to enhance the recognition accuracy. With samples from the BiosecurID database, we have analysed the entropy of each kind of trajectories, as well as the amount of information they share, and the difference between intra- and inter-writer measures of the mutual information. The results show that when pressure is not taken into account, the amount of information is similar in both types of trajectories. Furthermore, even if they share some information, in-air and on-surface trajectories appear to be notably non-redundant.     

Incremental Adaptive Learning Vector Quantization for Character Recognition with Continuous Style Adaptation

Incremental learning enables continuous model adaptation based on a constantly arriving data stream. It is a way relevant to human cognitive system, which learns to predict objects in a changing world. Incremental learning for character recognition is a typical scenario that characters appear sequentially and the font/writing style changes irregularly. In the paper, we investigate how to classify characters incrementally (i.e., input patterns appear once at a time). A reasonable assumption is that adjacent characters from the same font or the same writer share the same style in a short period while style variation occurs in characters printed by different fonts or written by different persons during a long period. The challenging issue here is how to take advantage of the local style consistency and adapt to the continuous style variation as well incrementally. For this purpose, we propose a continuous incremental adaptive learning vector quantization (CIALVQ) method, which incrementally learns a self-adaptive style transfer matrix for mapping input patterns from style-conscious space onto style-free space. After style transformation, this problem is casted into a common character recognition task and an incremental learning vector quantization (ILVQ) classifier is used. In this framework, we consider two learning modes: supervised incremental learning and active incremental learning. In the latter mode, samples receiving low confidence from the classifier are requested class labels. We evaluated the classification performance of CIALVQ in two scenarios, interleaved test-then-train and style-specific classification on NIST hand-printed data sets. The results show that local style consistency improves the accuracies of both two test scenarios, and for both supervised and active incremental learning modes.     

On the modularity of face recognition: The riddle of domain specificity

Abstract

The present paper focuses on the modular attributes of face recognition, defined in terms of domain specificity. Domain specificity is examined by looking into the innate nature of face recognition, the special effects related to the recognition of inverted faces, the specificity of electrophysiological responsivity to facial stimuli, and the specific impairment in face recognition associated with localized brain damage. Converging evidence from these sources seems to consistently show that face recognition is not qualitatively unique, as it proceeds in a manner similar to the recognition of other visuospatial objects. However, it seems to be special in that it may involve specific mechanisms dedicated to face recognition. Among infants, differential responsivity to faces and to other objects in terms of age of onset, attraction and course of development, seems to indicate the operation of a special process. Unusual inversion effects in face recognition might be due to the special expertise that humans develop for recognizing upright faces. Face-selective single unit responses in the monkey's brain implies the existence in the visual system of cells which are exclusively dedicated to the processing of facial stimuli. Finally, in prosopagnosia localized brain damage is linked to a specific inability to recognize familiar faces. Taken together, the data seem to show that some elements in the process of face recognition are domain specific, and in that sense, modular.     

Early visual ERP sensitivity to the species and animacy of faces

Assessing the agency of potential actors in the visual world is a critically important aspect of social cognition. Adult observers are generally capable of distinguishing real faces from artificial faces (even allowing for recent advances in graphics technology and motion capture); even small deviations from real facial appearance can lead to profound effects on face recognition. Presently, we examined how early components of visual event-related potentials (ERPs) are affected by the “life” in human faces and animal faces. We presented participants with real and artificial faces of humans and dogs, and analyzed the response properties of the P100 and the N170 as a function of stimulus appearance and task (species categorization vs. animacy categorization). The P100 exhibited sensitivity to face species and animacy. We found that the N170’s differential responses to human faces vs. dog faces depended on the task participants’ performed. Also, the effect of species was only evident for real faces of humans and dogs, failing to obtain with artificial faces. These results suggest that face animacy does modulate early components of visual ERPs—the N170 is not merely a crude face detector, but reflects the tuning of the visual system to natural face appearance.     

Covert recognition and the neural system for face processing

In this viewpoint, we discuss the new evidence on covert face recognition in prosopagnosia presented by Bobes et al. (2003, this issue) and by Sperber and Spinnler (2003, this issue). Contrary to earlier hypotheses, both papers agree that covert and overt face recognition are based on the same mechanism. In line with this suggestion, an analysis of reported cases with prosopagnosia indicates that a degree of successful encoding of facial representations is a prerequisite for covert recognition to occur. While we agree with this general conclusion as far as Bobes et al.'s and Sperber and Spinnler's data are concerned, we also discuss evidence for a dissociation between different measures of covert recognition. Specifically, studies in patients with Capgras delusion and patients with prosopagnosia suggest that skin conductance and behavioural indexes of covert face recognition are mediated by partially different mechanisms. We also discuss implications of the new data for models of normal face recognition that have been successful in simulating covert recognition phenomena (e.g., Young and Burton, 1999, and O'Reilly et al., 1999). Finally, in reviewing recent neurophysiological and brain imaging evidence concerning the neural system for face processing, we argue that the relationship between ERP components (specifically, N170, N250r, and N400) and different cognitive processes in face recognition is beginning to emerge.