In order to present a systematic overview of the topic, we have divided this article into several parts. In Section 2, we will describe the works related to our study. Section 3 gives the active shape model methods in details and 3D extend.
In section 4, we will present an overview of the application of image tracking. The conclusions and future works are given in Chapter 5.
Chapter 2
Related works
In this section, we review the related work about the face tracking in two part, face detection and tracking method.
2.1 Face detection
Before we tracking face, the most important process is to detect the face in an image. Here, we summarize method is in four classification to explain how to extract feature point in object and recent works . Here summarizes methods and representative works for face detection in a single image as shown in Table 2.1.
(A) Feature invariant approaches:[47]
This approach aims to find structural features that exist even when the pose, viewpoint, or lighting conditions vary, and then use the structural features to locate faces. These methods are designed mainly for face localization. Facial features such as eyebrows, eyes, nose, mouth are commonly extracted using edge detectors [7].
Researcher try to find the features of faces for detection. Leung et al. [30] use
Figure 2.1: Categorization of methods for face detection in a single image.
probability to locate a face from a chaotic scene based on local feature detectors.
Their idea is to find the arrangement of certain facial features that is most likely to be a face pattern. Yow and Cipolla [51] presented a feature based method that uses a large amount of samples from the visual image and their contextual samples.
They use Gaussian filter to obtain the points they interest, and then define the measurement on these point. An image region becomes a valid facial feature candi-date if the Mahalanobis distance between the corresponding feature vectors is below a threshold. Subsequently, this approach has been enhanced with active contour models by M. Kass et al. [24]. Augusteijn and Skufca [1] developed a method that deduces the presence of a face through the identification of face-like textures. They use second order statistical features to compute the measurement of textures on the sub-images of 16 × 16 pixels [21]. Y. Dai et al.[15] used a cascade correlation neural network to classify three types of features: skin, hair, and others.
(B) Template matching methods:
In template matching, a standard frontal face pattern is manually predefined or parameterized by a function. Sakai et al.[40] attempt to detect frontal faces in photographs. They set the sub-templates on features of face. Each sub-template is defined in terms of line segments which extracted based on greatest gradient change and then matched against the sub-templates, and we obtained the candidate po-sitions from their correlation. Craw et al. [13] presented a localization method based on a shape template of a frontal view face. After that, Yuille et al. [52] used deformable templates to model facial features that fit an a priori elastic model to facial features. Deformable templates is defined by the minimum energy function to link the edge of image. Later, this conception was extended to the method ”Snake”
proposed by M. Kass et al. [24]. Lam and Yan [31] used snakes to locate the head boundaries with a greedy algorithm in minimizing the energy function. Lanitis et al.
[32] use point distribution model and active shape model to represent image shapes and intensity information. Cootes and Taylor [20] applied a similar approach to localize a face in an image. They define rectangular regions of the image containing instances of the feature of interest. Cootes and Taylor applied a similar approach to localize a face in an image by rectangular regions of the image containing instances of the feature of interest [11].
(C) Appearance-based methods:
Appearance-based methods rely on techniques from statistical analysis and machine learning to find the relevant feature of face and non-face images. An early case to use eigenvectors in face recognition was done by Kohonen [28]. He proposed
a simple neural network to perform face recognition for aligned and normalized face images. These eigenvectors are known as eigenfaces. After that, Turk and Pentland [46] applied principal component analysis [22] to face recognition and detection.
Sung and Poggio [43] proposed distribution-based system for face detection. This system consists of distribution-based models for face or non-face patterns and a multi-inductor classifier. Neural networks have been applied successfully in many pattern recognition problems, such as optical character recognition, object recog-nition, and autonomous robot driving. Agui et al. [2] used hierarchical neural networks. They used parallel sub-networks in which the inputs are intensity values from an original image and intensity values from Sobel filter on image. Sobel fil-ter is used in image processing, particularly within edge detection algorithms. The first used Support Vector Machine to face detection was worked by Osuna et al.
[35] support vector machines re supervised learning models with associated learning algorithms that analyze data and recognize patterns, used for classification and re-gression analysis (Quote from Wikipedia). The basic SVM takes a set of input data and predicts, for each given input, which of two possible classes forms the input, making it a non-probabilistic binary linear classifier. Given a set of training exam-ples, each marked as belonging to one of two categories, an SVM training algorithm builds a model that assigns new examples into one category or the other. An SVM model is a representation of the examples as points in space, mapped so that the examples of the separate categories are divided by a clear gap that is as wide as possible. New examples are then mapped into that same space and predicted to belong to a category based on which side of the gap they fall on.