Optical character recognition (OCR) algorithms typically start from a binary label image. When presented with a grey scale image, ad-hoc methods, either global or local adaptive thresholding, are usually employed to c...
详细信息
ISBN:
(纸本)0819418455
Optical character recognition (OCR) algorithms typically start from a binary label image. When presented with a grey scale image, ad-hoc methods, either global or local adaptive thresholding, are usually employed to convert the grey scale image into a binary image. Such techniques have several serious deficiencies. A single global threshold may not exist that allows the extraction of all relevant label details. Even if a global threshold value does exist it may be difficult to determine. The local adaptive methods based upon empirical rules, work well for some labels, but are difficult to justify. Further the effect of the local adaptive thresholding algorithms is very dependent upon the choice of the window size. When inspecting packages moving on a conveyor belt we have control over the optical parameters of the system. Via autofocus and controlled lighting, parameters such as the optical path length, field of view, and illumination intensity may be adjusted. However no control can be placed on labels. The label reading system is totally subject to the package sender's whimsy. This paper describes the development of a recurrent neural network to segment grey scale label images into binary label images. To determine a pixel label, the neural network takes into account three sources of information: pixel intensities, correlations between neighboring labels, and edge gradients. These three sources of information are succinctly combined via the network's energy function. By changing its label state to minimize the energy function, the network satisfies constraints imposed by the input image and the current label values. To be mappable to analog hardware, it is desirable that the neural equations be deterministic. Two deterministic networks are developed and compared. The first operates at the zero temperature limit, the original Hopfield network. The second employs the mean field annealing algorithm. It is shown that with only a moderate increase in computational requ
neuralnetworks have been used to classify high resolution remote-sensed data. Experiments have demonstrated the potential of neuralnetworks for clustering large number of ground cover instances using supervised meth...
详细信息
neuralnetworks have been used to classify high resolution remote-sensed data. Experiments have demonstrated the potential of neuralnetworks for clustering large number of ground cover instances using supervised methods. This paper will describe a new algorithm of unsupervised learning, based on artificialneural network. Its performance has been compared with the competitive learning algorithm. The efficiency of this approach has been demonstrated through experimental results obtained on real-world of multispectral remote sensing data.
This paper describes a generalized fuzzy learning machine, which is a generalized and modified type of the neo-fuzzy-neuron presented by the authors in 1992. This machine can well grasp the nonlinear correlation of ea...
详细信息
ISBN:
(纸本)0818673125
This paper describes a generalized fuzzy learning machine, which is a generalized and modified type of the neo-fuzzy-neuron presented by the authors in 1992. This machine can well grasp the nonlinear correlation of each input. It has a very high nonlinear mapping ability compared with the conventional neuralnetworks, and it guarantees the global minimum. Furthermore, learning speed and its accuracy are improved drastically. It was successfully applied to the identification of the nonlinear dynamical system, e.g. two dimensional Lorenz chaotic model, and to the automatic detection of landmark location in the roentgenographic cephalogram for orthodontic treatment. The results were promising.
Texture has found many applications in computer vision. Examples where texture analysis methods are being used include: (i) classifying images and browsing images based on their texture;(ii) segmenting an input image ...
详细信息
Singular Value Decomposition, which has previously been applied to the problem of signal extraction from marine data, Zalloum et al (1), is currently being implemented for land use classification, Danaher et al (2). N...
详细信息
Singular Value Decomposition, which has previously been applied to the problem of signal extraction from marine data, Zalloum et al (1), is currently being implemented for land use classification, Danaher et al (2). neuralnetworks are becoming increasingly popular for the characterisation of multispectral remote sensing data, Benediktsson et al (3). There are a number of significant problems with using ANN as classifiers for this type of data. A comparison of these two procedures is performed and there merits and difficulties discussed. The authors introduce the Levenberg Marquardt technique as an advanced method for finding the global minima during a Backpropagation (BP) training scenario. These techniques are applied to simulated data generated from Landsat TM and SPOT satellite data of the County Wicklow area of Ireland. This data comprises five classes including Sitka spruce and Scots pine.
The proceedings contains 112 papers. Topics discussed include application of neuralnetworks in system identification, VLSI methodology, fading channel communications, electric power distribution, harmonics, nonlinear...
详细信息
The proceedings contains 112 papers. Topics discussed include application of neuralnetworks in system identification, VLSI methodology, fading channel communications, electric power distribution, harmonics, nonlinear control systems, microwave amplifiers, computer vision, artificial intelligence, discrete event systems, visual computing and communications, telecommunication networks, digital signal processing, waveguides, parallel processing, imageprocessing, rectifiers and inverters, speech processing, microwave measurements, multimedia and software programming, current control methods, microwave components and systems.
A convolution neural network (CNN) and a backpropagation neural network (BPN) were used for classification of regions of interest (ROIs) on mammograms as either mass or normal tissue. Input images to the CNN were obta...
详细信息
The proceedings contains 203 papers. Topics discussed include left ventricular function by magnetic resonance imaging, electrocardiography, cellular modeling, ventricular wall motion, body surface potential modeling, ...
详细信息
The proceedings contains 203 papers. Topics discussed include left ventricular function by magnetic resonance imaging, electrocardiography, cellular modeling, ventricular wall motion, body surface potential modeling, heart rate variability, information systems and databases, coronary angiography, arrhythmias, neuralnetworks, blood pressure and baroreflex control, coronary ultrasound, models of cardiac mechanics, models of electrical activation, electrophysiology, arteriography, imageprocessing, perfusion, computerized patient records, teaching, cardiac imaging, image communication, vascular modeling, ischemia, echocardiography, applications of artificial intelligence, myocardial activation and fibrillation.
A two-dimensional convolution neural network (CNN) with wavelet kernels (WK) has been developed for image pattern recognition. The structure of the CNN is a simplified version of the neocognitron. We used only a two-l...
详细信息
ISBN:
(纸本)0819417823
A two-dimensional convolution neural network (CNN) with wavelet kernels (WK) has been developed for image pattern recognition. The structure of the CNN is a simplified version of the neocognitron. We used only a two-level structure and eliminated all complex-cell layers. Nets between two adjacent layers in the feature selection level of the CNN are selectively interconnected across groups. In this part of the CNN signals processing, each group in the receiving layer receives signals from a group of weights (i.e., kernels). For the forward signal propagation, the product obtained from the kernel convoluting the front layer is collected onto the corresponding matrix element of the receiving layer. In this paper, the convolution kernels of the CNN (CNN/WK) are wavelet based and are trained by a supervised manner. In the development of the CNN/WK, we forced each updated convolution kernel to be orthonormal. Therefore, features (transformed coefficients) selected on the transform domain are linearly independent. Hence, the fully connected layers in the classification level of the CNN can perform more effectively. The applications of the CNN for disease pattern recognition have been very successful. When isolated patterns were further processed by internal filtering and classification layers were built into the neural network structure, the disease patterns were more easily recognized. Although, we did not receive substantial improvement of the ROC performance using the CNN/WK, this method may assist us in the analysis of the trained kernels and eventually lead to the optimization of feature extraction in a course of disease pattern recognition.
The proceedings contains 100 papers. Topics discussed include coupled filters for radars, launch vehicles, inverted pendulum, manipulators, shape measurement, artificial intelligence, velocity, flow and mass-flow meas...
详细信息
The proceedings contains 100 papers. Topics discussed include coupled filters for radars, launch vehicles, inverted pendulum, manipulators, shape measurement, artificial intelligence, velocity, flow and mass-flow measurement, signal processing, nonlinear system identification, adaptive control, three term control systems, robust control applications, neuralnetworks, optimization, Petri nets and fuzzy logic, active suspension, image reconstruction, optimal control, biomedical measurement, motion control, robot planning, guidance and control of aerospace, plant operation and diagnosing systems, traffic control, genetic algorithms, digital control and mobile robots.
暂无评论