In 2011, Ajit Danti have
proposed human face recognition of still images using face detection by
AdaBoost face detector, region of interest (ROI) extraction, feature extraction
using discrete wavelet transform (DWT), dimensionality reduction by employing
independent component analysis (ICA) and classification using k-Nearest
Neighborhood (k-NN) classifier.
Zahra Sadri Tabatabaie presents
a hybrid face detection system using a combination of appearance-based and
feature-based methods. They have combined Viola and Jones face detection method
with a color-based method to propose an improved face detection method.
Maity has suggested a novel approach to face detection using image parsing and
morphological analysis. The main objective of the paper as mentioned by the
authors is to propose an algorithm for extraction of some fundamental information
of an image efficiently and then finally use that to detect the human face on
Thakur has proposed a face detection method using skin tone segmentation. They
have proposed an algorithm which ingeniously uses a novel skin color model, RGB-HS-CbCr
for the detection of human faces.
Ahmad has discussed the various challenges in the area of image-based face
detection and recognition. As shown in the paper, Haar-like features reported
relatively well but it has much false detection than LBP which could be
considered being a future work in surveillance to reduce false detection in
Haar-like features and for the recognition part gabor is reported well as it’s
qualities overcomes datasets complexity.
To solve various modalities
issues such as the visual modality, the near infrared modality, and the sketch
modality Xiangsheng Huang, Zhen Lei, Mingyu Fan, Xiao Wang, and Stan Z. Li
suggests discriminative spectral regression (DSR) model for achieving robust
classification by mapping heterogeneous face images into a common
and Pentland describe a detection system based on principal component analysis
(PCA) subspace or eigen face representation. whereas only likelihood in the PCA
subspace is considered in the basic PCA method.
and Pentland also consider the likelihood in the orthogonal complement
subspace; using that system, the likelihood in the image space (the union of
the two subspaces) is modeled as the product of the two likelihood estimates,
which provide a more accurate likelihood estimate for the detection.
and Poggio first partition the image space into several face and nonface
clusters and then further decompose each cluster into the PCA and null subspaces.
The Bayesian estimation is then applied to obtain useful statistical features.
system of Rowley uses retinally connected neural networks. Through a sliding
window, the input image is examined after going through an extensive
train a nonlinear support vector machine to classify face and nonface patterns,
uses the SNoW (Sparse Network of Winnows) learning architecture for face
detection. In these systems, a bootstrap algorithm is used iteratively to
collect meaningful non-face examples from images that do not contain any faces
for retraining the detector.
and Kanade use multi resolution information for different levels of wavelet
transform. A nonlinear classifier is constructed using statistics of products of
histograms computed from face and nonface examples.
Nonlinear subspace methods
use nonlinear transforms to convert a face image into a feature vector in a
discriminative feature space. Kernel PCA 88 and kernel LDA 89 use kernel
tricks to map the original data into a high-dimension space to make the data
separable. Manifold learning, which assumes that face images occupy a
low-dimensional manifold in the original space, attempts to model such manifolds.
These include ISOMAP 90, LLE 91, and LPP 92. Although these methods achieve
good performance on the training data, they tend to over fit and hence do not
generalize well to unseen data.