Face detection is fundamental for several important appli- cations, such as, face recognition and human-computer interaction. This paper details a novel hierarchical face detector which combines motion, colour and late fusion. Motion segmentation is employed to eliminate background clutter and to reduce the initial search space. Subsequently, skin segmentation is used to determine a candidate face. Five simple eye-detection algorithms are fused to robustly localise the eyes within the candidate. Fusion is beneficial because the large amount of variation within the face makes it difficult for any individual technique to perform well under all conditions. The resulting system is capable of localising faces from still images in real-time with an accuracy of 93.75%.

/pdf/robust-face-localisation-using-motion-colour-fusion-4cswdslgkm.pdf

Robust Face Localisation Using Motion, Colour & Fusion

In this paper, the signal recognition by using quantum neural network (QNN) is studied and simulated. The signals with fuzziness distributed in the boundary of two different types of signals could be effectively recognized due to the structure of QNN's hidden units. To demonstrate the capability of QNN in recognition, the signals in a two-dimension (NC2) non-convex system is simulated. All the experiments are also performed by using the traditional neural network (NN) for a comparison.

Quantum NN vs. NN in signal recognition

In this paper, the non-stationary power signal prediction by using quantum neural network (QNN) is proposed. The signals with fuzziness are expected to be classified clearly for enhancing the learning efficiency of neural network due to the hidden units with various graded levels in QNN structure. For a comparison, all experiments are also performed using the conventional neural network (CNN) structure.

The non-stationary signal prediction by using quantum NN

A system for measuring the awareness of a subject including an imaging device adapted to acquire data representative of an image of the subject, and an awareness processor connected to the imaging device, wherein the awareness processor is adapted to classify the awareness of the subject according to the position of the subject.

System and method to determine awareness

Robust Face Localisation Using Motion, Colour and Fusion.

Proposes a fuzzy set theory to preprocess the feature of a Chinese character and uses the multi-moment method to classify the patterns in the neural network. Since it can be used in overlapping classes, this method has practical value.

A multi-moment pattern recognition method based on neural networks

A face recognition method is provided based on a facial action reconstructing method. Traditional face recognition methods treated different expressions of a face as some kind of noises whereas various expressions were treated as a rule of the facial actions in our method. Based on two knowledge increasable model conditions we presented, different facial expressions are reconstructed with an optical flow method and then applied as a training set to achieve single sample face recognition. Furthermore, these samples are whitened using an eigenvalue decomposition method. Our face recognition is based on ICA representations. A better recognition rate is achieved this way. This kind of face expression reconstruction method can be used in image compression and facial animation based on MPEG-4.

Face recognition: a facial action reconstruction and ICA representation approach

Research the small sample face recognition approach. A small sample statistics face recognition approach based on virtual sample is presented. In this method, prototype faces are used to create virtual sample to extend the sample set, independent component analysis (ICA) algorithm is used to reduce dimensions and extract features, and then Bayesian expectation maximisation algorithm is adopted to learn the parameters of condition probability function of class features. A Bayes classifier is constructed to recognize test samples. Experience result shows the approach can improve face recognize rate of small sample efficiently.

Face recognition method based on virtual sample

The paper presents a dynamic grouping multi-class face recognition method, which has knowledge-increasable ability and can solve the problems of large classes face recognition and pattern classes dynamic extension. By adopting multiple classifiers parallel working in the process of training and dynamic grouping recognition, the method can not only speed up calculation and improve the recognition rate but also achieve extension easily and freely. Experimental results prove its reasonableness and feasibility.

http://ieeexplore.ieee.org/iel5/8405/26504/01180020.pdf

Dynamic grouping multi-classes face recognition method

An extendable framework is developed for an ensemble HME model based on the theoretical analysis of information geometry. In a hierarchical set of systems, a lower order system is included in the parameter space of a large one as a subset. Such a parameter space has rich geometrical structures that are responsible for the dynamic behavior of learning. The HME network divides a task into small tasks by the principle of divide and conquer to improve the performance of a single network. By studying the dual manifold architecture for mixtures of neural networks and analyzing the probability of knowledge-increasable model based on information geometry, the paper proposes a new method to achieve the multi-HME model that has knowledge-increasable and structure-extendible ability. The method helps to provide explanation of the transformation mechanism of the human recognition system and understand the theory of the global architecture of neural network.

Wen Jin-wei

Papers

A multi-moment pattern recognition method based on neural networks

Face recognition: a facial action reconstruction and ICA representation approach

Face recognition method based on virtual sample

Dynamic grouping multi-classes face recognition method

Information geometry on ensemble HME model