Browsing by Author "YEH, BF"

Now showing 1 - 7 of 7

A COLOR CONSTANCY MODEL FOR ADVANCED TELEVISION CAMERAS
(1992-06-01)
CHANG, PR
;
HSIEH, TH
;
YEH, BF
;
交大名義發表
;
電信工程研究所
In high-fidelity color sensing, one may attempt to use recorded video image data from an electronic camcorder to generate a rendering that appears the same as the original image did at the time of acquisition. To achieve the goal, a color constancy mechanism is proposed to perceive the image color from the strengths of three RGB responses of a color camera or the like, independent of the color of the light illuminating the object. Being able to extract the color descriptors that are independent of the illumination is desirable because of the varity of the situation in which color is important but illumination conditions cannot be controlled. The first step of our method is to use a finite dimensional linear model to estimate the color signals. We have shown that any color signal can be characterized as a linear combination of four principal component basis functions. Once the color signals have been estimated, the unknown illuminant can be determined by the Tominaga and Wandell's estimation method which is based on a dichromatic model. In color constancy, the knowledge of surface reflectance helps in determining the canonical color descriptors despite the variation in the spectral power distribution of the ambient light. It is shown that the estimate of surface reflectance is derived from both the estimated color signal and illuminant straightforwardly. The model of a combination of the estimations of the surface spectral reflectance and the color signals can be employed in designing a color constancy electronic video camera which indeed improves the shortcomings of the traditional video camcorders.
A NUMERICALLY STABLE PIPELINE NET VLSI ARCHITECTURE FOR THE ISOMORPHIC HOPFIELD MODEL
(1993-05-01)
CHANG, PR
;
YEH, BF
;
交大名義發表
;
電信工程研究所
;
National Chiao Tung University
This correspondence presents a reconfigurable pipeline net VLSI architecture for implementing Hopfield neural models. It is known that the Hopfield models involve computing the hyperbolic trigonometric functions which are hard to be realized by digital VLSI architectures. In order to tackle such difficulty, a useful isomorphic nonlinear mapping is introduced to convert those hyperbolic trigonometric nonlinear functions into the simple second-order polynomial functions. Moreover, the isomorphic formulation provides the higher ability to decompose the problem into several independent tasks which can be assigned to a number of processors. Handling the digital realizations on the Hopfield model, the previous attemps were to use the technique based on the first-order approximation called Euler's method which has poor numerical stability and large truncation error. To find a numerical solution with a prescribed accuracy, one of the promising approaches is a combination of both a single-step Runge-Kutta method and a multistep predictor-corrector method which has a larger stability interval and is particularly suitable for parallel computation. Since the mixed-type procedure requires data broadcasting, common VLSI architectures with fixed connections cannot offer such flexible connectivities. A pipeline net VLSI architecture which is a programmable two-level pipelined and dynamically reconfigurable systolic array would be adopted as the design platform. The pipelining period and block pipelining period of the proposed architecture have the computational orders of O(1) and O(n), respectively, where n is the number of neurons.
ADAPTIVE PACKET EQUALIZATION FOR INDOOR RADIO CHANNEL USING MULTILAYER NEURAL NETWORKS
(1994-08-01)
CHANG, PR
;
YEH, BF
;
CHANG, CC
;
交大名義發表
;
電信工程研究所
This paper investigates the application of the multilayer perceptron structure to the packet-wise adaptive decision feedback equalization of a M-ary QAM signal through a TDMA indoor radio channel in the presence of intersymbol interference (ISI) and additive Gaussian noise. Since the multilayer neural networks are capable of producing complex decision regions with arbitrarily nonlinear boundaries, this would greatly improve the performance of conventional decision feedback equalizer (DFE) where the decision boundaries of conventional DFE are linear. However, the applications of the traditional multilayer neural networks have been limited to real-valued signals. To tackle this difficulty, a neural-based DFE is proposed to deal with the complex QAM signal over the complex-valued fading multipath radio channel without performing time-consuming complex-valued back-propagation training algorithms, while maintaining almost the same computational complexity as the original real-valued training algorithm. Moreover, this neural-based DFE trained by packet-wise backpropagation algorithm would approach an ideal equalizer after receiving a sufficient number of packets. In this paper, another fast packet-wise training algorithm with better convergence properties is derived on the basis of a recursive least-squares (RLS) routine. Results show that the neural-based DFE trained by both algorithms provides a superior bit-error-rate performance relative to the conventional least mean square (LMS) DFE, especially in poor signal to noise ratio conditions.
INVERSE FILTERING OF A LOUDSPEAKER AND ROOM ACOUSTICS USING TIME-DELAY NEURAL NETWORKS
(1994-06-01)
CHANG, PR
;
LIN, CG
;
YEH, BF
;
交大名義發表
;
電信工程研究所
This paper presents the design of a neural-based acoustic control used for the equalization of the response of a sound reproduction system. The system usually can be modeled as a composite system of a loudspeaker and an acoustic signal-transmission channel. Generally, an acoustic signal radiated inside a room is linearly distorted by wall reflections. However, in a loudspeaker, the nonlinearity in the suspension system produces a significant distortion at low frequencies and the inhomogeneity in the flux density causes a nonlinear distortion at large output signals. Both the linear and nonlinear distortions should be reduced so that high fidelity sound can be reproduced However, the traditional adaptive equalizer which is only capable of dealing with linear systems or specific nonlinear systems cannot compensate these nonlinear distortions. The time-delay feedforward neural network (TDNN) which has the capability to learn arbitrary nonlinearity and process the temporal audio patterns are particularly recognized as the best nonlinear inverse filter of the composite system. The performance of a TDNN-based acoustic controller is verified by some simulation results.
MODEL-REFERENCE COLOR REPRODUCTION FOR VIDEO SYSTEM BASED ON BACKPROPAGATION NETWORKS
(1994)
CHANG, PR
;
TAI, CC
;
YEH, BF
;
電信工程研究所
;
Institute of Communications Engineering
NEURAL-BASED PACKET EQUALIZATION FOR INDOOR RADIO CHANNEL BY FAST BACK-PROPAGATION ALGORITHM
(1994)
CHANG, PR
;
YEH, BF
;
TSENG, CM
;
電信工程研究所
;
Institute of Communications Engineering
NONLINEAR COMMUNICATION CHANNEL EQUALIZATION USING WAVELET NEURAL NETWORKS
(1994)
CHANG, PR
;
YEH, BF
;
電信工程研究所
;
Institute of Communications Engineering