To effectively represent facial features in complex environments, a face recognition method based on dense grid histograms of oriented gradients (HOG) is proposed. First, the face image is divided by numerous dense grids from which the HOG features are extracted. Then, all the grid HOG feature vectors are composed to realize the feature expression of the whole face, and the nearest neighbor classifier is used for recognition. In the FERET face database with complex changes in illumination, time and environment, we test the gamma illumination correction, the spatial gradient direction, the size of the block, the standardization, and the face image resolution to find and analyze the optimal HOG parameters for face recognition. Finally, we compare our dense grid HOG with the two famous local facial feature extraction methods: the Gabor wavelet and the local binary pattern (LBP) on face recognition. The experimental results show that the dense grid HOG method is more suitable for the variations in time and environment. The feature extraction times of the dense grid HOG and LBP are similar. However, the dense grid HOG method uses fewer dimensions to obtain a better recognition rate than LBP. Moreover, the dense grid HOG feature extraction time greatly outperforms the Gabor wavelet feature.
25 July 2019 in IEEE Journal of Biomedical and Health Informatics
The recently developed multiscale-based fusion methods can be improved with two approaches: an advanced image decomposition scheme and an advanced fusion rule. In this paper, three-layer image decomposition, enhanced illumination fusion rule-based fusion method is proposed. The proposed method includes three steps. First, each input image is decomposed into its corresponding smooth, texture and edge layers using defined local extrema and low-pass filters in the spatial domain. Second, three different strategies are applied as fusion rules for the three-layer representation. To preserve the illumination closely related to tumors, the illumination is corrected by applying a higher contrast to the decomposed image details, including the texture and edge inputs, such as those found in grayscale CT and MRI images. The final fused image is created by the addition of the normalized smooth, texture and edge image layers. The experiments demonstrate that the proposed method performs better than the existing state-of-the-art fusion methods.
Image set classification has attracted increasing attention with respect to the use of significant amounts of within-set information. The covariance matrix is a natural and effective descriptor for describing image sets. Non-singular covariance matrices, known as symmetric positive definite (SPD) matrices, are regarded as points on a Riemannian manifold. A common method of classifying points on a manifold is to explicitly map the SPD matrices from a Riemannian manifold to Euclidean space, such as in the covariance discriminative learning (CDL) method. However, the disadvantages of the CDL method are as follows: (1) the method models the whole image set as a covariance matrix, whereas if there are insufficient set samples or merely one set, the within-class information studied by the discriminative learning may not be utilized well and (2) when the original sample covariance matrices are of high dimensionality, the computational cost is non-trivial. To address these problems of CDL, we propose to exploit the maximal linear patch to cluster image sets into multiple subsets (local patches), which could provide substantially more within-class information. Moreover, we refine the manifold formed by the SPD matrices to a lower dimensionality and more discriminative manifold by collaboratively applying principal component analysis to all training sets. Experiments are performed on face recognition and objection categorization tasks; extensive comparison results illustrate the considerable effectiveness of our proposed method.
Currently, neighborhood linear discriminant analysis (nLDA) exploits reverse nearest neighbors (RNN) to avoid the assumption of linear discriminant analysis (LDA) that all samples from the same class should be independently and identically distributed (i.i.d.). nLDA performs well when a dataset contains multimodal classes. However, in complex pattern recognition tasks, such as visual classification, the complex appearance variations caused by deformation, illumination and visual angle often generate non-linearity. Furthermore, it is not easy to separate the multimodal classes in lower-dimensional feature space. One solution to these problems is to map the feature to a higher-dimensional feature space for discriminant learning. Hence, in this paper, we employ kernel functions to map the original data to a higher-dimensional feature space, where the nonlinear multimodal classes can be better classified. We give the details of the deduction of the proposed kernel reverse neighborhood discriminant analysis (KRNDA) with the kernel tricks. The proposed KRNDA outperforms the original nLDA on most datasets of the UCI benchmark database. In high-dimensional visual recognition tasks of handwritten digit recognition, object categorization and face recognition, our KRNDA achieves the best recognition results compared to several sophisticated LDA-based discriminators.
Set-based image classification has become popular in recent years since it can provide a relatively large amount of within-set information that benefits classification. Grassmann Discriminant Analysis (GDA) models image sets as points (subspaces) on a Grassmann manifold and then explicitly maps them to a higher-dimensional Hilbert space, where Euclidean geometry applies, for Linear Discriminant Analysis (LDA). However, due to the noise disturbance and finite number of training sammples in practice, the conventional problems of LDA, such as the singularity of the within-class scatter matrix and the instability of its inverse, also appear with GDA, which result in recognition performance deterioration. Inspired by eigenspectrum regularization techniques, we propose an eigenspectrum Regularized GDA (RGDA) method to alleviate the conventional problems of GDA in Grassmannian space. Moreover, we implement it with the graph embedded framework, three different eigenspectrum regularization models are incorporated into the proposed approach respectively. Extensive experimental results on set-based face recognition and object categorization tasks have confirmed the effectiveness of our approaches.