A novel face recognition approach based on strings of minimum values and several distance metrics

Linear discriminant analysis (LDA) is a well-known method for face recognition in feature extraction and dimension reduction. As a new scheme, two-dimensional linear discriminant analysis (2DLDA) has been used for face recognition recently. In this paper, an assembled matrix distance metric based 2DLDA is proposed for face representation and recognition. In this new method, an assembled matrix distance (AMD) metric is used to measure the distance between two 2DLDA feature matrices. To test this new method, ORL face database is used and the results show that the assembled matrix distance metric based 2DLDA method outperforms the 2DLDA method and achieves higher classification accuracy than the 2DLDA algorithm

There are many distance metrics used in nearest neighbor classifiers. In this paper a grey relation metric, the balance incident degree (BID) is proposed as distance metric based on exploring the theory about grey relational analysis. The presented approach with the method of eigenface is applied in face recognition and compared with the common distance metrics. Through the experiments with ORL dataset, the recognition rates for BID are better than other metrics in the same condition. BID method can also cope with the high dimension problem. The results demonstrate BID can be a good distance metric and deduce into more applications.

The nearest-neighbor (NN) classifier has long been used in pattern recognition, exploratory data analysis, and data mining problems. A vital consideration in obtaining good results with this technique is the choice of distance function, and correspondingly which features to consider when computing distances between samples. In this paper, a new ensemble technique is proposed to improve the performance of NN classifier. The proposed approach combines multiple NN classifiers, where each classifier uses a different distance function and potentially a different set of features (feature vector). These feature vectors are determined for each distance metric using Simple Voting Scheme incorporated in Tabu Search (TS). The proposed ensemble classifier with different distance metrics and different feature vectors (TS-DF/NN) is evaluated using various benchmark data sets from UCI Machine Learning Repository. Results have indicated a significant increase in the performance when compared with various well-known classifiers. Furthermore, the proposed ensemble method is also compared with ensemble classifier using different distance metrics but with same feature vector (with or without Feature Selection (FS)).

In this paper, a theoretically efficient method is developed for face recognition. It is based on two dimensional principal component (2DPCA) analysis and extended local binary pattern (Extended LBP, ELBP) texture. First, the ELBP operator is employed to extract the local texture of the face images. Second, 2DPCA is used to reduce the dimensionality of the extracted feature and get the optimal projection space. Finally, the nearest distance classification is used to distinguish each testing image. The method has been ac-cessed on ATR-Jaffe and AR face databases. Results demonstrate that the proposed method is obviously superior to PCA and 2DPCA, and its recognition rate is more stable than PCA. Meanwhile, the proposed method has strong robustness against illumination and facial expression changes. INTRODUCTION In the field of biometrics recognition, face recognition has good application prospects in the field of identification, security monitoring and human-computer interaction. Factors affecting the face recognition mainly lie in: the changeable facial expressions and pose [11]; the same face will change with age; two dimensional face images are susceptible to be affected by illumination when they are filmed; recognizing a specific face involves a lot of knowledge, such as image preprocessing , pattern recognition and computer vision. Feature extraction is key to face recognition. Global feature extraction methods include principal component analysis (PCA) [9], independent component analysis (ICA) [6] and linear discriminate analysis (LDA) [12] under the conditions that samples obey the multivariate normal distribution. The advantages of PCA method is to get global features which can represent the characteristics of the face image by K-L transform, the disadvantage is that two-dimensional image matrix need to be transformed into one-dimensional vector, resulting in a huge amount of computation. Meanwhile PCA has a bad robustness under the illumination, pose and facial expressions. Jian Yang developed a novel two-dimensional PCA (2DPCA) [10] method. Compared with global features, local feature has a strong robustness against the changes of illumination and pose. The common local feature extraction methods include local binary pattern (LBP) [2, 8], scale invariant feature transform (SIFT) and so on. LBP is an effective texture extraction method and has been extensively exploited in many applications, for instance, image analysis, texture classification [1, 5, 7], environment modeling and other fields [3]. Classifiers commonly used in face recognition are: minimum distance method, nearest neighbor classifier, support vector machine classifier and BP neural network method. Based on 2DPCA, this paper proposed a fusion feature extraction method of 2DPCA and extended LBP (ELBP) texture, in the stage of classification a reliable algorithm—the nearest classifier—is utilized to identify a specific face image, these measures effectively reduce the influence of illumination, facial expressions and pose changes and improve the recognition rate. FUSION FEATURE EXTRACTION OF 2DPCA AND EXTENDED LBP The proposed method This paper aimed to improve face recognition rate under lighting changes and facial expressions. 2nd International Conference on Electrical, Computer Engineering and Electronics (ICECEE 2015) © 2015. The authors Published by Atlantis Press 1398 Figure 1 illustrates the flow chart of the improved method. Figure 1. The structure diagram of face recognition of the proposed method First, preprocess images to enhance information. Second, use LBP to describe faces. Third, obtain the optimal projection space by 2DPCA. Fourthly, project test set and training set into the optimal projection space and obtain a feature matrix for each image. Finally, employ a classifier to classify, summarize the recognition rate and do some analysis comparing with other similar algorithms. LBP face appearance descriptor LBP-based feature extraction operator is computationally efficient statistical characteristics that can distinguish different objects in the same image in microcosm form [6], and is not sensitive to changes of gray-scale and different lighting conditions. LBP operator represents texture by comparing the gray value of the center pixel with the gray-scale of its neighboring pixels. It has been proved a powerful approach to describe local structure [5]. The LBP pattern of each pixel is defined as:

We concern the problem of learning a Mahalanobis distance metric for improving nearest neighbor classification. Our work is built upon the large margin nearest neighbor (LMNN) classification framework. Due to the semidefiniteness constraint in the optimization problem of LMNN, it is not scalable in terms of the dimensionality of the input data. The original LMNN solver partially alleviates this problem by adopting alternating projection methods instead of standard interior-point methods. Still, at each iteration, the computation complexity is at least O(D3) (D is the dimension of input data). In this work, we propose a column generation based algorithm to solve the LMNN optimization problem much more efficiently. Our algorithm is much more scalable in tha tat each iteration, it does not need full eigen-decomposition. Instead, we only need to find the leading eigen value and its corresponding eigen vector, which is of O(D2) complexity. Experiments show the efficiency and efficacy of our algorithms.

Marginal Fisher analysis (MFA) is a representative margin-based learning algorithm for face recognition. A major problem in MFA is how to select appropriate parameters, k 1 and k 2, to construct the respective intrinsic and penalty graphs. In this paper, we propose a novel method called nearest-neighbor (NN) classifier motivated marginal discriminant projections (NN-MDP). Motivated by the NN classifier, NN-MDP seeks a few projection vectors to prevent data samples from being wrongly categorized. Like MFA, NN-MDP can characterize the compactness and separability of samples simultaneously. Moreover, in contrast to MFA, NN-MDP can actively construct the intrinsic graph and penalty graph without unknown parameters. Experimental results on the ORL, Yale, and FERET face databases show that NN-MDP not only avoids the intractability, and high expense of neighborhood parameter selection, but is also more applicable to face recognition with NN classifier than other methods.

Object-oriented classification is a useful tool for analysis of high-resolution imagery due to the incorporation of spectral, textural and contextual variables. However, feature selection and incorporation of appropriate training sites can be difficult. We compared two object-oriented image classification approaches, one using a decision tree (DT), the other a nearest neighbor classification (NN) with regard to classification accuracy, effort involved and feasibility for mapping similar areas. We used a QuickBird satellite image to map arid rangeland vegetation in a 1200 ha pasture in southern New Mexico. In the DT approach, we used ground truth data from plots (8.75 m) as input for a decision tree to create a rule base for classification. In the NN approach, larger polygons (mean=100 m) served as training areas for a nearest neighbor classification. Overall accuracy was 80% using the DT and 77% using the NN classification. The DT was a superior tool for reducing the number of input features, but this technique required more field data, export to a decision tree program and was more time consuming. With the NN approach, input features were selected within the image analysis program and were applied to the classification immediately. The use of larger polygons for training and test samples was more appropriate for use in an object-oriented environment than the small plots. We concluded that for arid rangeland classification from QuickBird data, the NN technique required less time in the field and for image analysis, had comparable accuracy to the DT approach, and would be appropriate for mapping similar areas. A combination of both methods would incorporate the advantages of feature selection in a DT with the object-oriented nature of the analysis.

A set-to-set nearest neighbor approach for robust and efficient face recognition with image sets

A nearest neighbor classifier based on virtual test samples for face recognition

Multi-hypothesis nearest-neighbor classifier based on class-conditional weighted distance metric

In this paper we study how to improve nearest neighbor classification by learning a Mahalanobis distance metric. We build on a recently proposed framework for distance metric learning known as large margin nearest neighbor (LMNN) classification. Our paper makes three contributions. First, we describe a highly efficient solver for the particular instance of semidefinite programming that arises in LMNN classification; our solver can handle problems with billions of large margin constraints in a few hours. Second, we show how to reduce both training and testing times using metric ball trees; the speedups from ball trees are further magnified by learning low dimensional representations of the input space. Third, we show how to learn different Mahalanobis distance metrics in different parts of the input space. For large data sets, the use of locally adaptive distance metrics leads to even lower error rates.

Improving nearest neighbor classification with cam weighted distance

Due to its wide applications in practice, face recognition has been an active research topic. With the availability of adequate training samples, many machine learning methods could yield high face recognition accuracy. However, under the circumstance of inadequate training samples, especially the extreme case of having only a single training sample, face recognition becomes challenging. How to deal with conflicting concerns of the small sample size and high dimensionality in one-sample face recognition is critical for its achievable recognition accuracy and feasibility in practice. Being different from the conventional methods for global face recognition based on generalization ability promotion and local face recognition depending on image segmentation, a single-sample face recognition algorithm based on locality preserving projection (LPP) feature transfer is proposed here. First, transfer sources are screened to obtain the selective sample source using the whitened cosine similarity metric. Second, we project the vectors of source faces and target faces into feature subspace by LPP, respectively, and calculate the feature transfer matrix to approximate the mapping relationship on source faces and target faces in subspace. Then, the feature transfer matrix is used on training samples to transfer the original macro characteristics to target macro characteristics. Finally, the nearest neighbor classifier is used for face recognition. Our results based on popular databases FERET, ORL, and Yale demonstrate the superiority of the proposed LPP feature transfer-based one-sample face recognition algorithm when compared with popular single-sample face recognition algorithms, such as (PC) $^{2}\text{A}$ and Block FLDA.

Modern Tracking algorithms treat tracking as a binary classification problem between the object class and the background class. In this paper, we propose the use of Distance Metric Learning (DML) in combination with Nearest Neighbor (NN) classification for object tracking. Initially a video file is read and the frames in the video are accessed individually. The object in that video is first detected using canny edge detector. We assume that the previous appearances of the object and the background are clustered so that a nearest neighbor classifier can be used to distinguish between the new appearance of the object and the appearance of the background. Using Nearest Neighbor classifier it is able to distinguish the object from other objects. The process is repeated for all the frames. Then the object is tracked using the Distance Metric Learning algorithm using normalized correlation between the frames. The human appearance model is identified using the Blob detector which uses the skin color to identify the object. Then the bounding box is fixed for the object in that frame. Then the video is reconstructed with the processed frames. Feature extraction is done using Region Props which threshold the image and extract the features. Measure the gray level co-occurrence matrix and match the best similar one.

Super-Resolution (SR) involves the registration of multiple images/frames and reconstruction of a single higher resolution image. The goal of this research is to use multip le, very-low resolution images, such as those produced from a video sequence in a wireless sensor network system, as input to the super-resolution process in a face recognition system. The algorithm used for face reco gnition is the Fisherfaces method with a n earest neighbor classifier used for the recognition decision. Super-resolution consists of two stages, a registration stage and a reconstruction stage. Testing images were segmented using a simple skin color detection approach. After cropping they were combined into groups of four to be used for the super-resolution algorithm using faces from three people. Each group of four images was used as an input to the Keren registration algorithm where the rotational and translation information was saved that was then entered into the robust super-resolution reconstruction algorithm to create a single high quality image, which was processed by the face recognition algorithm. An av erage of the same groups of four was tested as well as a centroid shifted average. Comparison was based on n earest neighbor classifier and on classification rates. The results were not in favor of the super-resolution method but instead, the centroid shifted average was the best in this study. Keywords: Face Recognition, Super-Resolution, Fish erfaces, Nearest Neighbor Classifier