Kuang-Chiung Chang

Comparison of similarity measures for clustering electrocardiogram complexes

This paper compares four similarity measures such as the city block (L1-norm), the Euclidean (L2-norm), the normalized correlation coefficient, and the simplified gray relational grade for clustering QRS complexes. Performances of the measures include classification accuracy, threshold value selection, noise robustness, and execution time. The clustering algorithm used is the so-called two-step unsupervised method. The best out of the 10 independent runs of the clustering algorithm with randomly selected initial template beat for each run is used to compare the performances of each similarity measure. Simulation results show that the simplified gray relational grade outperforms the other measures

A comparison of similarity measures for clustering of QRS complexes

Similarity or distance measures play important role in the performance of algorithms for ECG clustering problems. This paper compares four similarity measures such as the city block (L1-norm), Euclidean (L2-norm), normalized correlation coefficient, and simplified grey relational grade for clustering of QRS complexes. Performances of the measures include classification accuracy, threshold value selection, noise robustness, execution time, and the capability of automated selection of templates. The clustering algorithm used is the so-called two-step unsupervised method. The best out of the 10 independent runs of the clustering algorithm with randomly selected initial template beat for each run is used to compare the performances of each similarity measure. To investigate the capability of automated selection of templates for ECG classification algorithms, we use the cluster centers generated by the clustering algorithm with various measures as templates. Four sets of templates are obtained, each set for a measure. And the four sets of templates are used in the k-nearest neighbor classification method to evaluate the performance of the templates. Tested with MIT/BIH arrhythmia data, we observe that the simplified grey relational grade outperforms the other measures in classification accuracy, threshold value selection, noise robustness, and the capability of automated selection of templates.

Modeling and Detection of Blurred Illumination Edges

Luminance edges are classified into reflectance edges and illumination edges. Most of the edge detection papers published focus on reflectance edges detections, whereas only few of them deal with illumination edges. In this paper, a new approach to detect illumination edges in grayscale images is proposed. Differing from the conventional illumination edge detection techniques, which were mostly based on the grayscale intensity changes in the image being processed, instead, an error-function-based formulation is introduced to model blurred illumination edges, using a factor called the illumination ratio defined in this paper. Then, the profile of the illumination ratio is fitted with the proposed error-function model to determine the locations of all illumination edges. Although the frame-differencing method using complementary patterns is generally acknowledged to be the most accurate illumination edge detector, the proposed technique could perform better than it in both speed and accuracy. This technique benefits the fields of visual detection, machine vision, and especially the coded structured-light-based 3D scanning systems.

TV settop box design for processing text messages based on graph theory

This paper proposes a TV settop box design for processing text messages (captions). With this settop box, a user could catch the text messages in which the user is interested from a TV program, such as news or sports, while watching a program coming from another source. The settop box extracts text messages from the TV program and pastes the extracted results onto frames of the program being watched before they are actually displayed on the screen. Although text messages are composed by characters, they are not recognized by any means. Instead, text messages are extracted and treated as bitmap images. A graph-theory- based algorithm was implemented and tested in the settop box to process text messages in real-time. This algorithm could correctly and efficiently process text messages and most importantly, could be applied on messages in any language without modifications.

CLUSTERING ECG COMPLEXES USING SELF- ORGANIZING CMAC

In the study, we apply a self-organizing cerebellar model articulation controller (SOCMAC) network to cluster QRS complexes. Forty-four out of 48 records from the MIT-BIH Arrhythmia Database are used to test the SOCMAC network. The SOCMAC network is an unsupervised and nonlinear learning method. For each record, under fixed initial memory contents, we search optimal parameters in the updating rule of the network for achieving the best classification accuracy. The entire classification accuracy, with the optimal parameters, is 99.41%, which is higher than existing results. The testing data are divided into two groups: noisy and clean. The classification accuracy of each group differs not much. Therefore, we can conclude that the SOCMAC shows a certain degree of robustness against noise.

Conversion of 3D Triangular-Meshed Model to Dot Representation

Dot representation is the only format accepted in a special 3D printing application called laser stippling, which outputs the shape and texture tone of a 3D model inside a lead glass by using laser beam. To fully utilize existing abundant and exquisite 3D triangular-meshed models, this study proposes a two-phase technique, i.e., a sampling phase followed by a halftoning phase, to convert models in triangular-meshed format to dot representation. The sampling phase generates a set of dense sample points that cover the entire surface of the model, and the halftoning phase removes some of the sample points to reflect the tone of the model’s texture. The actual stippled results demonstrated the technique could deliver products in good quality.

Design of smart volume controller for televisions

This paper presents a rule-based algorithm for designing a smart volume controller used in TVs. The algorithm is able to detect variations of loudness coming from different audio sources, i.e., TV programs and commercials, and adjust the volume controller intelligently so that the maximum allowable loudness is kept while maintaining the dynamic range of the programs. The design has been implemented and tested. Experimental results show the effectiveness of this design.