Canonical Correlation Analysis Algorithm

Published Date: 02 Nov 2017

Abstract

In video surveillance, the faces of interest are often of small size. Image resolution is an important factor affecting face recognition by human and computer. In existing system a feature extraction method called coupled kernel embedding (CKE) is used for LR face recognition. The final kernel matrix is constructed by concatenating two individual kernel matrices in the diagonal direction. CKE solves this problem by minimizing the dissimilarities captured by their kernel Gram matrices. In the proposed system canonical correlation analysis (CCA) algorithm, also called Local discrimination CCA (LDCCA) along with coupled kernel embedding (CKE) is used for LR & HR face image recognition. LDCCA method considers a combination of local properties and discrimination between classes. Not only combination between the sample pairs but also the correlation between the samples and local neighborhoods are taken into consideration. CCA can extract the features even though there is a wound in the face . In the proposed system the time complexity will be reduced. Automatic face recognition has long been established as one of the most active research areas in computer vision

Face recognition in unconstrained environments remains challenging for most practical applications.

Keywords Face Recognition, Feature Extraction, CKE, CCA, Low Resolution, Super Resolution.

1. Introduction

A database consists of static images of human faces. Images were taken in uncontrolled indoor environment using cameras of various qualities. Automatic face recognition has long been established as one of the most active research areas in computer vision. Face recognition in unconstrained environments remains challenging for most practical applications. In contrast to traditional still-image based approaches, recently the research focus has shifted towards video based approaches. Video data provides rich and redundant information, which can be exploited to resolve the inherent ambiguities of image-based recognition like sensitivity to low resolution, pose variations and occlusion, leading to more accurate and robust recognition. Face recognition has also been considered in the content-based video retrieval setup, for example, character-based video search.

Another feasible solution adopting the smoothing and downsampling strategy has been proposed in the literature.We can get lower resolution images using smoothing and downsampling implementations on the HR image database; then we use the two modes to train the projection directions, where the LR samples are mainly used to determine optimal metric while the HR samples are exploited for providing

The A video camera produces images at a certain frame rate and depth of field, which impose physical limits on the spatial density of image detectors. Intuitively, recovering the lost information of LR face images first is a promising solution for achieving better performance. In fact, most existing two-step methods of LR face recognition exploit a preprocessing of SR as the first step. The super-resolved face images are then passed to the second step for recognition and classification. During the past decade, a number of learning-based SR methods have been proposed to predict the corresponding HR image from a single LR image or multiple LR images. Almost all the learning-based methods consider a linear projection subspace to solve the problem of computing similarity metrics between HR and LR images projection subspace to solve the problem of computing similarity metrics between HR and LR images computing similarity metrics between HR and LR images projection subspace to solve the problem of computing similarity metrics between HR and LR images. A new efficient kernel method for LR faces recognition without any SR preprocessing. According to the aim of recognition, learn a coupled kernel embedding (CKE) method to map the face images with different resolutions onto an infinite subspace and carry out the recognition step in the new space. Human motion tracking is primarily concerned with determining the existence and location of humans within certain regions of space. Automatic face recognition is a process of identifying a test face image with one of the faces stored in a prepared face database. Video analytics, also called intelligent video surveillance, is a technology that uses software to automatically identify specific objects, behaviors or attitudes in video footage. It transforms the video into data to be transmitted or archived so that the video surveillance system can act accordingly. It may involve activating a mobile camera in order to obtain more specific data about the scene or simply to send a warning to surveillance personnel so that a decision may be made on the proper intervention required.

2. Related Works

During the past decade, a number of learning-based SR methods have been proposed to predict the corresponding HR image from a single LR image or multiple LR images. Baker and Kanade propose “face hallucination” to infer the HR face image from an input LR face image based on face priors. Chang et al. propose a method based on locality linear embedding. Jia and Gong propose a multi linear approach to hallucinate face images across multiple modalities (generalization to variations such as facial expression or pose) based on a unified global and local tensor space representation. Li et al. solve SR reconstruction using a sparse directional regularization strategy for color images. Recently, several algorithms have been proposed to avoid explicit SR in the image domain. Henningsâ€"Yeomans et al. propose a joint objective function that integrates the aims of SR and face recognition, which indeed improves the recognition rate. Zou and Yuen discover the nonlinear relationship between the LR subspace and HR subspace with a kernel regression (KR) approach. The KR method consists of two phases, namely, training and recognition. In the training phase, given a set of LR and HR image pairs, both images are mapped to the kernel feature space by nonlinear mappings. In most surveillance scenarios there is a large distance between the camera and the objects of interest in the scene. Surveillance cameras are also usually set up with wide fields of view in order to image as much of the scene as possible. The end result is that the objects in the scene normally appear very small in surveillance imagery.

To analyze the super-resolution reconstruction constraints, In particular, derive a sequence of results which all show that the constraints provide far less useful information as the magnification factor increases. Therefore propose an algorithm that learns recognition-based priors for specific classes of scenes, the use of which gives far better super-resolution results for both faces and text. Generally, all super-resolution algorithms are based on the fundamental constraints that the super-resolution image should generate the low resolution input images when appropriately warped and down-sampled to model the image formation process. A face recognition algorithm which is insensitive to large variation in lighting direction and facial expression. Taking a pattern classification approach, consider each pixel in an image as a coordinate in a high-dimensional space. Take advantage of the observation that the images of a particular face, under varying illumination but fixed pose, lie in a 3D linear subspace of the high dimensional image space if the face is a Lambertian surface without shadowing The Fisher faces method, a derivative of Fisher’s Linear Discriminate (FLD), maximizes the ratio of between-class scatter to that of within-class scatter.

The Eigen face method is also based on linearly projecting the image space to a low dimensional feature space. However, the Eigen face method, which uses principal components analysis (PCA) for dimensionality reduction, yields projection directions that maximize the total scatter across all classes. Computational analysis shows that SR has only linear-time complexity which is a huge speed up comparing to the cubic-time complexity of the ordinary approaches. Experimental results on face recognition demonstrate the effectiveness and efficiency of method, Propose a new regularized subspace learning framework called Spectral Regression (SR).

The framework is developed from a graph embedding viewpoint of dimensionality reduction algorithms. Super-resolution through neighbor embedding proposes a novel method for solving single-image super-resolution problems. Given a low-resolution image as input, recover its high resolution counterpart using a set of training examples. While this formulation resembles other learning based methods for super-resolution, this method has been inspired by recent manifold learning methods, particularly locally linear embedding (LLE). Specifically, small image patches in the low- and high-resolution images form manifolds with similar local geometry in two distinct feature spaces. This approach simultaneously provides measures of fit of the super-resolution result, from both reconstruction and recognition perspectives. Face recognition degrades when faces are of very low resolution since many details about the difference between one person and another can only be captured in images of sufficient resolution. In this work, a new procedure for recognition of low-resolution faces, when there’s a high-resolution training set available

3. Techniques Used

3.1 Coupled Kernel Embedding (CKE)

Kernel-based learning machines have aroused considerable interest in the fields of pattern recognition and machine learning. Kernel representation offers an alternative learning to nonlinear functions by projecting the data onto a high-dimensional Hilbert feature space to increase the computational power of the linear learning machines, though this still leaves unresolved the issue of how best to choose the features or the kernel function in ways that will improve performance.

𝚽˸𝑹𝑴→𝑽 𝒙→𝚽(ð")

Let 𝚽 be a nonlinear mapping, and the HR input data space 𝑹𝑴 is mapped onto a (potentially much higher dimensional) feature vector in the feature space V.

CKE algorithm implemented on a binary classification problem. There are two modes including HR objects (squares and five-pointed stars) and their LR counterparts (triangles and circles) in the input space, where the squares and triangles belong to the first class, and the five-pointed stars and circles belong to the second class. It is well known that the nonlinear kernel mapping could transform complex distributed data onto high-dimensional.

Hilbert feature space where the data becomes linearly separable. Moreover, the kernel method and the inner product operation can be efficiently used to represent the nonlinear features in the reproducing kernel Hilbert space . The data points in the original individual input spaces onto different reproducible kernel Hilbert spaces by coupled nonlinear functions ( fsr and gsr), i.e., At the same time, the samples coming from different modes or different classes may become more separable in the kernel spaces. Some special properties (e.g., isomorphic) may be endued to the Hilbert spaces V and W.

Ψ:𝑹m→w x→ψ(x)

In the same way, let Ψ be another nonlinear mapping corresponding to LR images, and the input data 𝑹m space with the other mode (m << M) can be mapped onto the feature space w. Then determine the embedding features using a locality-preserving projections (LPP) approach and implement the recognition stage in the learned embedding subspace by some classification procedure. It is worth noting that the target of implicit SR step is consistent with classification since the kernel mappings will be integrated well into a uniform objective function using the locality-preserving criterion, and the algorithm is very efficient and suitable for real-time applications.

3.2 Canonical Correlation Analysis Algorithm (CCA)

A new feature extraction algorithm is developed based on canonical correlation analysis (CCA), called Local Discrimination CCA (LDCCA). The method considers a combination of local properties and discrimination between different classes. Not only the correlations between sample pairs but also the correlations between samples and their local neighborhoods are taken into consideration in LDCCA.

Effective class separation is achieved by maximizing local within-class correlations and minimizing local between-class correlations simultaneously. Besides, a kernel version of LDCCA (KLDCCA) is proposed to cope with nonlinear problems in experiments. The experimental results on an artificial dataset, multiple feature databases and face databases including ORL, Yale, AR validate the effectiveness of the proposed methods.

Canonical correlation analysis (CCA), just like Principal component analysis is an effective feature extraction method for dimensionality reduction and data visualization. PCA is a single-modal method, which deals with data samples obtained from a single information channel or view. In contrast, CCA is typically used for multi-view data samples. In order to improve the performance of CCA in classification tasks, so incorporate the idea of local discriminate analysis into CCA, which is referred to as LDCCA.

A facial recognition system is a computer application for automatically identifying or verifying a person from a digital image or a video frame from a video source. One of the ways to do this is by comparing selected facial features from the image and a facial database. It is typically used in security systems and can be compared to other biometrics such as fingerprint or eye iris recognition systems.

4. Conclusion and Future Works

This paper studied the impact of the number of learning based SR methods have been proposed to predict the corresponding HR image from a single LR image or multiple LR images. While the earlier existing algorithms are time consuming. Moreover these are not suitable for real-time application. In the proposed system Canonical Correlation Analysis (CCA) algorithm along with coupled kernel embedding (CKE) is used for LR and HR face image recognition.

Semisupervised learning problems are the focus of our future work.