Chin-Chuan Han

Face Recognition Using Nearest Feature Space Embedding

Face recognition algorithms often have to solve problems such as facial pose, illumination, and expression (PIE). To reduce the impacts, many researchers have been trying to find the best discriminant transformation in eigenspaces, either linear or nonlinear, to obtain better recognition results. Various researchers have also designed novel matching algorithms to reduce the PIE effects. In this study, a nearest feature space embedding (called NFS embedding) algorithm is proposed for face recognition. The distance between a point and the nearest feature line (NFL) or the NFS is embedded in the transformation through the discriminant analysis. Three factors, including class separability, neighborhood structure preservation, and NFS measurement, were considered to find the most effective and discriminating transformation in eigenspaces. The proposed method was evaluated by several benchmark databases and compared with several state-of-the-art algorithms. According to the compared results, the proposed method outperformed the other algorithms.

Hyperspectral Image Classification Using Nearest Feature Line Embedding Approach

Eigenspace projection methods are widely used for feature extraction from hyperspectral images (HSI) for the classification of land cover. Projection transformation is used to reduce higher dimensional feature vectors to lower dimensional vectors for more accurate classification of land cover types. In this paper, a nearest feature line embedding (NFLE) transformation is proposed for the dimension reduction (DR) of an HSI. The NFL measurement is embedded in the transformation during the discriminant analysis phase, instead of the matching phase. Three factors, including class separability, neighborhood structure preservation, and NFL measurement, are considered simultaneously to determine an effective and discriminating transformation in the eigenspaces for land cover classification. Three state-of-the-art classifiers, the nearest-neighbor, support vector machine, and NFL classifiers, were used to classify the reduced features. The proposed NFLE transformation is compared with different feature extraction approaches and evaluated using two benchmark data sets, the MASTER set at Au-Ku and the AVIRIS set at Northwest Tippecanoe County. The experimental results demonstrate that the NFLE approach is effective for DR in land cover classification in the field of Earth remote sensing.

Greedy modular eigenspaces and positive Boolean function for supervised hyperspectral image classification

This paper presents a new supervised classification tech- nique for hyperspectral imagery, which consists of two algorithms, re- ferred to as the greedy modular eigenspace (GME) and the positive Boolean function (PBF). The GME makes use of the data correlation matrix to reorder spectral bands from which a group of feature eigen- spaces can be generated to reduce dimensionality. It can be imple- mented as a feature extractor to generate a particular feature eigens- pace for each of the material classes present in hyperspectral data. The residual reconstruction errors (RREs) are then calculated by projecting the samples into different individual GME-generated modular eigens- paces. The PBF is a stack filter built by using the binary RRE as classi- fier parameters for supervised training. It implements theminimum clas- sification error (MCE) as a criterion so as to improve classification performance. Experimental results demonstrate that the proposed GME feature extractor suits the nonlinear PBF-based multiclass classifier well for classification preprocessing. Compared to the conventionalprincipal components analysis (PCA), it not only significantly increases the accu- racy of image classification but also dramatically improves the eigende- composition computational complexity.

Data fusion of hyperspectral and SAR images

A novel technique is proposed for data fusion of earth remote sensing. The method is developed for land cover classification based on fusion of remote sensing images of the same scene collected from multiple sources. It presents a framework for fusion of multisource remote sensing images, which consists of two algorithms, referred to as the greedy modular eigenspace (GME) and the feature scale uniformity transformation (FSUT). The GME method is designed to extract features by a simple and efficient GME feature module, while the FSUT is performed to fuse most correlated features from different data sources. Finally, an optimal positive Boolean function based multiclass classifier is further developed for classification. It utilizes the positive and negative sample learning ability of the minimum classification error criteria to improve classification accuracy. The performance of the proposed method is evaluated by fusing MODIS/ASTER airborne simulator (MASTER) images and the airborne synthetic aperture radar (SAR) images for land cover classification during the PacRim II campaign. Experimental results demonstrate that the proposed fusion approach is an effective method for land cover classification in earth remote sensing, and improves the precision of image classification significantly compared to conventional single source classification.

A GPU-Based Simulation of Tsunami Propagation and Inundation

Tsunami simulation consists of fluid dynamics, numerical computations, and visualization techniques. Nonlinear shallow water equations are often used to model the tsunami propagation. By adding the friction slope to the conservation of momentum, it also can model the tsunami inundation. To solve these equations, we use the second order finite difference MacCormack method. Since it is a finite difference method, it brings the possibility to be parallelized. We use the parallelism provided by GPU to speed up the computations. By loading data as textures in GPU memory, the computation processes can be written as shader programs and the operations will be done by GPU in parallel. The results show that with the help of GPU, the simulation can get a significant improvement in the execution time for each of the computation steps.

Vehicle color classification using manifold learning methods from urban surveillance videos

Color identification of vehicles plays a significant role in crime detection. In this study, a novel scheme for the color identification of vehicles is proposed using the locating algorithm of regions of interest (ROIs) as well as the color histogram features from still images. A coarse-to-fine strategy was adopted to efficiently locate the ROIs for various vehicle types. Red patch labeling, geometrical-rule filtering, and a texture-based classifier were cascaded to locate the valid ROIs. A color space fusion together with a dimension reduction scheme was designed for color classification. Color histograms in ROIs were extracted and classified by a trained classifier. Seven different classes of color were identified in this work. Experiments were conducted to show the performance of the proposed method. The average rates of ROI location and color classification were 98.45% and 88.18%, respectively. Moreover, the classification efficiency of the proposed method was up to 18 frames per second.

The color identification of automobiles for video surveillance

Color identification of automobiles plays a significant in intelligent transportation systems (ITS). In this paper, a novel scheme for color identification of automobile is proposed using the taillight detection and a template matching module. The taillights of cars are detected to find the valid regions of interested (ROIs) for color identification. The color feature vectors generated by 3 by 3 neighboring pixels are classified by a template matching strategy. Seven classes, red, yellow, blue, green, black, white, and gray, are identified in this work. Experimental results have been conducted to show the validity of the proposed method. The averaged accuracy rate 81.71% is achieved and the performance of this scheme is up to 20 frames per second.

A modular eigen subspace scheme for high-dimensional data classification

In this paper, a novel filter-based greedy modular subspace (GMS) technique is proposed to improve the accuracy of high-dimensional data classification. The proposed approach initially divides the whole set of high-dimensional features into several arbitrary number of highly correlated subgroups by performing a greedy correlation matrix reordering transformation for each class. These GMS can be treated as not only a preprocess of GMS filter-based classifiers but also a unique feature extractor to generate a particular feature subspaces for each different class presented in high-dimensional data. The similarity measures are next calculated by projecting the samples into different modular feature subspaces. Finally, a GMS filter-based architecture based on the mean absolute errors criterion is adopted to build a non-linear multi-class classifier. The proposed GMS filter-based classification scheme is developed to find non-linear boundaries of different classes for high-dimensional data. It not only significantly improves the classification accuracy but also dramatically reduces the computational complexity of feature extraction compared with the conventional principal components analysis. Experimental results demonstrate that the proposed GMS feature extraction method suits the GMS filter-based classifier best as a classification preprocess. It significantly improves the precision of high-dimensional data classification.

Biased Discriminant Analysis With Feature Line Embedding for Relevance Feedback-Based Image Retrieval

The focus of content-based image retrieval (CBIR) is to narrow down the gap between low-level image features and high-level semantic concepts. In this paper, a biased discriminant analysis with feature line embedding (FLE-BDA) is proposed for performance enhancement in relevance feedback schemes. Maximizing the margin between relevant and irrelevant samples at local neighborhoods was the aim in this study. In reduced subspace, relevant images and query images can be quite close, while irrelevant samples are far away from relevant samples. The results of four benchmark datasets are given to show the performance of the proposed method.

Falling and slipping detection for pedestrians using a manifold learning approach

Falling activity is a critical behavior due to the physical discomfort for elders. The prime time of rescuing is missed whenever falls accidentally happen. Fall detection in real time could save human life in video surveillance systems. Recently, digital cameras are installed everywhere. Human activities are monitored from cameras by intelligent programs. An alarm is sent to the administrator when an abnormal event occurs. In this paper, a multi-view-based manifold learning algorithm is proposed for detecting the falling events. This algorithm should be able to detect people falling down in any direction. First, the walking patterns in a normal speed are modeled by the locality preserving projection (LPP). Since the duration of falling activity is hard to be estimated from real videos, partial temporal windows are matched with the normal walking patterns. The Hausdorff distances are calculated to estimate the similarity. In the experiments, the falling events are effectively detected by the proposed method.