Hwei-Jen Lin

Tamkang University, T’ai-pei, Taipei, Taiwan

Are you Hwei-Jen Lin?

Claim your profile

Publications (35)1.05 Total impact

  • Source
    Hwei-Jen Lin, Hung-Hsuan Wu, Chun-Wei Wang
    J. Inf. Sci. Eng. 01/2011; 27:95-110.
  • Source
    Journal of Multimedia. 01/2010; 5:522-527.
  • Hwei-Jen Lin, Jih Pin Yeh
    [Show abstract] [Hide abstract]
    ABSTRACT: The main issue is to search for a subset of the support vector solutions produced by an SVM that forms a discriminant function best approximating the original one. The work is accomplished by giving a fitness (objective function) that fairly indicates how well the discriminant function formed by a set of selected vectors approximates the original one, and searching for the set of vectors having the best fitness using PSO, EGA, or a hybrid approach combining PSO and EGA. Both the defined fitness function and the adopted search technique affect the performance. Our method can be applied to SVMs associated with any general kernel. The reduction rate can be adaptively adjusted based on the requirement of the task. The proposed approach is tested on some benchmark datasets. The experimental results show that the proposed method using PSO, EGA, or a hybrid strategy combining PSO and EGA associated with the objective function defined in the paper outperforms both the method proposed by Li et al. (2007) and our previously proposed method (Lin and Yeh, 2009), and that a hybrid strategy of PSO and EGA provides better results than a single strategy of PSO or EGA.
    Pattern Recognition Letters. 01/2010; 31:563-571.
  • Source
    Hwei-Jen Lin, Chung-Yung Chen, Hua-Wei Hsia
    [Show abstract] [Hide abstract]
    ABSTRACT: We propose a facial feature replacement system, which uses the triangulation algorithm to perform facial feature replacement in each segmented triangular region associated with control points. The experimental results show that our system provides quite natural composite images. In addition, the system is flexible and has no limit in the shape, size, and plane rotation of the faces which are processed.
    01/2010;
  • Hwei-Jen Lin, Li-Kang Chen, Chun-Wei Wang
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents a flexible and effective colorization system that carries out colorization by performing two stages of chrominance determination for pixels in a gray-scale image, including scribble region expansion and chrominance blending. The region corresponding to each scribble color is determined by expanding the scribbles in parallel associated with the edge map of the given image. Usually a few pixels remain unassigned to a color region after this stage. For those pixels we define and compute the transportation distance between each pixel and each scribble color. Then the reciprocals of the distances are used as weights to assign each of those pixels a weighted sum of the scribble colors. To promote practicability and effectiveness of the system several modules are provided that make the system more user-friendly and enhance the colorization results. The experimental results show that the proposed system is not only flexible and easy to use, but it also provides natural and satisfactory colorization results.
    Parallel Architectures, Algorithms, and Networks, International Symposium on. 12/2009;
  • Source
    Hwei-Jen Lin, Chun-Wei Wang, Yang-Ta Kao
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper proposes a method for detecting copy-move forgery over images tampered by copy-move. To detect such forgeries, the given image is divided into overlapping blocks of equal size, feature for each block is then extracted and represented as a vector, all the extracted feature vectors are then sorted using the radix sort. The difference (shift vector) of the positions of every pair of adjacent feature vectors in the sorting list is computed. The accumulated number of each of the shift vectors is evaluated. A large accumulated number is considered as possible presence of a duplicated region, and thus all the feature vectors corresponding to the shift vectors with large accumulated numbers are detected, whose corresponding blocks are then marked to form a tentative detected result. Finally, the medium filtering and connected component analysis are performed on the tentative detected result to obtain the final result. Compared with other methods, employing the radix sort makes the detection much more efficient without degradation of detection quality.
    WSEAS Transactions on Signal Processing 05/2009; 5(5):188-197.
  • Source
    Hwei-Jen Lin, I-Chun Pai, Fu-Wen Yang
    [Show abstract] [Hide abstract]
    ABSTRACT: Face Recognition is an important topic in the field of pattern recognition. This technology has a variety of applications including entrance guard control, personal service system, criminal verification, and security verification of finance. Our research focuses on the development of a human face recognition system. It is a challenge to correctly identify a human in an image under various possible situations including difference of lighting conditions, change of hairstyles, variation of facial expression, and different aspects of the face. We have analyzed several existing face recognition techniques and found that each of them is performed well over some specific sets of testing samples but poorly over some other sets. This motivates us to combine different techniques to construct a better face recognition system. First, we propose a new module E-2DPCA applying DCT for image enhancement and 2DPCA for feature extraction. The experimental results show that the recognition accuracy of E-2DPCA is better than all the modules we have analyzed. We choose the best two from those analyzed and compared them with our proposed E-2DPCA module, and found that although the E-2DPCA module outperforms the other two modules, each of the three modules behaves better than others over some specific set of samples. Thus we combine the three modules and apply weighted voting scheme to choose the recognition result from those given by the three modules. Experimental results show that the integrated system can further improve the recognition rate.
    Next-Generation Applied Intelligence, 22nd International Conference on Industrial, Engineering and Other Applications of Applied Intelligent Systems, IEA/AIE 2009, Tainan, Taiwan, June 24-27, 2009. Proceedings; 01/2009
  • Hwei-Jen Lin, Chen-Wei Chang, I-Chun Pai
    [Show abstract] [Hide abstract]
    ABSTRACT: The performance of face recognition systems depends on conditions being consistent, including lighting, pose and facial expression. To solve the problem produced by pose variation it is suggested to pre-estimate the pose orientation of the given head image before it is recognized. In this paper, we propose a head pose estimation method that is an improvement on the one proposed by N. Hu et al. The proposed method trains in a supervised manner a nonlinear interpolative mapping function that maps input images to predicted pose angles. This mapping function is a linear combination of some Radial Basis Functions (RBF). The experimental results show that our proposed method has a better performance than the method proposed by Nan Hu et al. in terms of both time efficiency and estimation accuracy.
    01/2009;
  • [Show abstract] [Hide abstract]
    ABSTRACT: For object detection and tracking, we use a modified version of Gaussian Mixture Models (GMMs) to construct the background, and then subtract it from the image to obtain the foreground where the moving objects are located. We then perform some operations, including shadow removal, edge detection, and connected component analysis to localize each moving object in the foreground. As soon as an object is detected, it is tracked in the subsequent frames using a Particle Filter (PF). The PF is effective, but the dimension of its state space is high since the tracked objects tend shift. To reduce this problem, we modify the particle filter by tracking over the foreground portion instead of the entire image. Using modified versions of both the GMM and PF, our system proves to have a high accuracy rate for detection/tracking and satisfactory time efficiency.
    Far East Journal of Experimental and Theoretical Artificial Intelligence. 01/2009; 3(2).
  • Hwei-Jen Lin, Jih Pin Yeh
    [Show abstract] [Hide abstract]
    ABSTRACT: Being a universal learning machine, a support vector machine (SVM) suffers from expensive computational cost in the test phase due to the large number of support vectors, and greatly impacts its practical use. To address this problem, we proposed an adaptive genetic algorithm to optimally reduce the solutions for an SVM by selecting vectors from the trained support vector solutions, such that the selected vectors best approximate the original discriminant function. Our method can be applied to SVMs using any general kernel. The size of the reduced set can be used adaptively based on the requirement of the tasks. As such the generalization/complexity trade-off can be controlled directly. The lower bound of the number of selected vectors required to recover the original discriminant function can also be determined.
    Applied Mathematics and Computation. 01/2009; 214:329-335.
  • Hwei-Jen Lin, Li-Kang Chen, Chun-Wei Wang
    The 10th International Symposium on Pervasive Systems, Algorithms, and Networks, ISPAN 2009, Kaohsiung, Taiwan, December 14-16, 2009; 01/2009
  • Shwu-Huey Yen, Wen-Tsung Tsai, Hwei-Jen Lin
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents a shot boundary detection algorithm using morphological opening with structuring elements consistent to the acute level of a shot. By this way, the difficulty on indistinctness of gradual transitions is solved. The algorithm is applied to soccer video and the resulting shots are further classified as long, medium, or close-up/others shots. The classification algorithm is based on grass pixel ratio which is calculated with motion and color features without the need of pre-training or updating. Both shot detection and classification algorithms are tested on various types of videos and soccer video with excellent results.
    Far East Journal of Experimental and Theoretical Artificial Intelligence. 01/2009; 3(2).
  • [Show abstract] [Hide abstract]
    ABSTRACT: This paper proposes an improved version of our previously introduced face detection system based on skin color segmentation and neural networks. The new system uses a support vector machine (SVM) based method for verification.
    Secure System Integration and Reliability Improvement, 2008. SSIRI '08. Second International Conference on; 08/2008
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: We propose a method to extract text information from video sequences. First the frequency of high horizontal energy in a video frame is examined to extract text blocks. Structural operations are then performed to remove the background so that the text can be extracted for later recognition. Experiments show that the method is efficient and effective for extracting text from various video documents.
    Secure System Integration and Reliability Improvement, 2008. SSIRI '08. Second International Conference on; 08/2008
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Abstract—In a mobile computing system, users carrying portable devices can access database services from any location without requiring a fixed position in the networking environ- ment. Some examples of strategies supported by databases in mobile computing include location dependent queries, long-lived transactions that require migration of data into the portable devices, form-based transactions, and online information report. Within a mobile computing environment, the need for a real- time database management model is strong, because one of the basic requirements in mobile data management,is to provide real-time response to transactions of the underlying strategies. However, the resource constraints of mobile computing systems make,it difficult to satisfy timing requirements of supported strategies. Low bandwidth, unreliable wireless links, and fre- quent disconnections increase the overhead of communication between mobile hosts and,fixed hosts of the system. There
    Journal of Software - JSW. 01/2008; 3(8):65-72.
  • Hwei-Jen Lin, Hung-Hsuan Wu
    [Show abstract] [Hide abstract]
    ABSTRACT: Let X[0..n-1] and Y[0..m-1] be two sorted arrays, and define the mxn matrix A by A[j][i]=X[i]+Y[j]. Frederickson and Johnson [G.N. Frederickson, D.B. Johnson, Generalized selection and ranking: Sorted matrices, SIAM J. Computing 13 (1984) 14-30] gave ...
    Information Processing Letters 01/2008; 109:116-120. · 0.49 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: A content-based painting image retrieval (CBPIR) system based on AdaBoost is proposed. By providing query examples which share the same semantic concepts, e.g., portraits, and incorporating with relevance feedback (RF), the user can acquire the desired painting images. To bridge the gap between low-level features and semantic concepts, a large set of 4,356 features on texture and spatial arrangement of painting images is provided. Utilize the nice characteristic of AdaBoost algorithm that it can combine partial weak classifiers, i.e. features, into a strong one, the system can correctly discover a few most critical features from provided samples and search paintings sharing same features from the database. Our experiment in query of "portrait," based on 3 RFs and an average of 50 repetitions, shows an excellent performance of (approximately) 0.71, 0.84, 0.95 in Precision, Recall, and Top 100 Precision rates. The average execution time, based on 50 repetitions, required in initial query and three RF with training and classifying is approximately 1.2 seconds, thus a complete query takes less than 5 seconds in training and classifying. The system is proved to be accurate in content based image retrieval and also very efficient for on-line users.
    Systems, Man and Cybernetics, 2006. SMC '06. IEEE International Conference on; 11/2006
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents a scheme to extract inscriptions from a traditional Chinese painting such that the inscriptions and the painting can be enjoyed or studied separately. A two phases morphological operation is used to remove most content of a painting (i.e. background) which makes inscriptions to become the principal object in the remaining image. Since inscriptions are written vertically, we use the alignment property to construct the center point map and use it to locate character lines. Character block is formed by clustering adjacent character lines. The proposed algorithm has been executed on a set of Chinese paintings and proved its efficacy.
    Systems, Man and Cybernetics, 2006. SMC '06. IEEE International Conference on; 11/2006
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, we propose an unsupervised genetic clustering algorithm, which produces a new chromosome without any conventional genetic operators, and instead according to the gene reproducing probabilities determined by Markov chain modeling. Selection of cluster centers from the dataset enables construction of a look-up table that saves the distances between all pairs of data points. The experimental results show that the proposed algorithm not only solves the premature problem to provide a more stable clustering performance in terms of number of clusters and clustering results, but also improves the time efficiency.
    18th International Conference on Pattern Recognition (ICPR 2006), 20-24 August 2006, Hong Kong, China; 01/2006
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper proposes a Content-Based Image Retrieval (CBIR) system applicable in mobile devices. Due to the fact that different queries to a content-based image retrieval (CBIR) system emphasize different subsets of a large collection of features, most CBIR systems using only a few features are therefore only suitable for retrieving certain types of images. In this research we combine a wide range of features, including edge information, texture energy, and the HSV color distributions, forming a feature space of up to 1053 dimensions, in which the system can search for features most desired by the user. Through a training process using the AdaBoost algorithm9 our system can efficiently search for important features in a large set of features, as indicated by the user, and effectively retrieve the images according to these features. The characteristics of the system meet the requirements of mobile devices for performing image retrieval. The experimental results show that the performance of the proposed system is sufficiently applicable for mobile devices to retrieve images from a huge database.
    International Journal of Pattern Recognition and Artificial Intelligence 01/2006; 20:525-542. · 0.56 Impact Factor