Figure 4 - uploaded by Hamparsum Bozdogan
Content may be subject to copyright.
CMC curve of recognition using the structurally diverse feature clusters learnt using our formulation.
Source publication
In this paper, we derive a data mining framework to analyze 3D features on human faces. The framework leverages kernel density estimators, genetic algorithm and an information complexity criterion to identify discriminant feature-clusters of lower dimensionality. We apply this framework on human face anthropometry data of 32 features collected from...
Context in source publication
Context 1
... conducted two experiments with the face data and our feature learning scheme. The first one was to learn feature clusters (combinations) that are minimal and informative by executing the GA to convergence. The GA pruned 200,000 feature clusters and the top 3 clusters with the lowest ICOMP values in the converged population were: (FC-1) width of nose, depth of nose, depth at the eye, angle at nose along its length (FC-2) depth at the eye, lip to chin distance, nose to lip distance, jaw to jaw distance (FC-3) vertical profile, length of eyebrow, length of nose. With these feature clusters we have essentially learnt what set of features to try and measure in situations where not all measurements of a sparse probe face are possible. The second experiment that we conducted was by reinitializing the population for the GA after a fixed number of iterations and observing the convergence over several trials. We saved the optimal result after every 1000 iterations and ranked features based on their frequency of occurrence in the converged set over 100 such trials. We observed that the vertical profile, jaw to jaw distance, depth of nose, depth at the eye and chin to neck distance features repeated with a high frequency greater than 60% while other features repeated less than 30% of the time. With the knowledge of minimal and informative features, our next experiment was to evaluate the recognition performance with these features. Our probes were a low resolution sparse point cloud of the face generated by decimating the dense 3D model to a mesh with only 100 3D vertices. This is the typical resolution of deformed generic meshes used in face modeling using video sequences. We used the Euclidean distance between features as our recognition metric to generate the results in Figure 4. 3D face recognition has faced criticism for its inability to handle expressions and for the co-operation required for success in real-world crime situations. Our study has helped us cluster facial features based on their discriminatory characteristics after studying a large-diverse database of human faces. Our conclusions explain the success of several heuristics experimented in the past for face recognition using only horizontal and vertical profiles and caricature inspired 3D face recognition. Our future efforts are towards constructing an n -point anthropometric 3D face-graph of expression invariant geometric features, with anthropometric point features as nodes and informative-discriminative geometric features as weighted attributes in the graph. We expect the graph matching to provide us a reliable expression-invariant 3D face recognition method extending the previous effort with 2D face images [14] to ...
Similar publications
Hyperspectral image visualization reduces color bands to three, but prevailing linear methods fail to address data characteristics, and nonlinear embeddings are computationally demanding. Qualitative evaluation of embedding is also lacking. We propose faithful stochastic proximity embedding (FSPE), which is a scalable and nonlinear dimensionality r...
We explore the use of several linear dimensionality reduction techniques
that can be easily integrated into the hyperspectral imaging sensor. We
investigate their effect on the performance of classical target
detection and classification techniques for hyperspectral images.
Specifically, each N-dimensional spectral pixel is embedded to an
M-dimensi...
Dimensionality reduction techniques are especially important in the context of embedded vision systems. A promising dimensionality reduction method for a use in such systems is the random projection. In this paper we explore the performance of the random projection method, which can be easily used in embedded cameras. Random projection is compared...
To verify cluster separation in high-dimensional data, analysts often reduce the data with a dimension reduction (DR) technique, and then visualize it with 2D Scatterplots, interactive 3D Scatterplots, or Scatterplot Matrices (SPLOMs). With the goal of providing guidance between these visual encoding choices, we conducted an empirical data study in...
Unsupervised regression is a dimensionality reduction method that allows embedding high-dimensional patterns in low-dimensional latent spaces. In the line of research on iterative unsupervised regression, numerous methodological variants have been proposed in the recent past. This works extends the set of methods by evolutionary embeddings. We prop...
Citations
Faces convey a wealth of social signals, including race, expression, identity, age and gender, all of which have attracted increasing attention from multi-disciplinary research, such as psychology, neuroscience, computer science, to name a few. Gleaned from recent advances in computer vision, computer graphics, and machine learning, computational intelligence based racial face analysis has been particularly popular due to its significant potential and broader impacts in extensive real-world applications, such as security and defense, surveillance, human computer interface (HCI), biometric-based identification, among others. These studies raise an important question: How implicit, non-declarative racial category can be conceptually modeled and quantitatively inferred from the face? Nevertheless, race classification is challenging due to its ambiguity and complexity depending on context and criteria. To address this challenge, recently, significant efforts have been reported toward race detection and categorization in the community. This survey provides a comprehensive and critical review of the state-of-the-art advances in face-race perception, principles, algorithms, and applications. We first discuss race perception problem formulation and motivation, while highlighting the conceptual potentials of racial face processing. Next, taxonomy of feature representational models, algorithms, performance and racial databases are presented with systematic discussions within the unified learning scenario. Finally, in order to stimulate future research in this field, we also highlight the major opportunities and challenges, as well as potentially important cross-cutting themes and research directions for the issue of learning race from face.
