Electrical Engineering

In this paper, a neural network (NN) based approach for classification of images represented by translation-, scale-, and rotation-invariant features is presented. The utilized network is a multilayer perceptron (MLP) classifier with one hidden layer. The backpropagation learning is used for its training. Two types of features are used: moment invariants derived from geometrical moments of the image, and the newly developed Zernike moment based features. Zernike moments are the mapping of the image onto a set of complex orthogonal polynomials. The performance of the MLP is compared to those of three other traditional statistical classifiers, namely, Bayes, nearestneighbor, and minimum-mean-distance. Through extensive experimentation with noiseless as well as noisy binary images of all English characters (26 classes), the following conclusions are reached: 1) the MLP outperforms the other three classifiers, especially when noise is present, 2) the nearest-neighbor classifier performs about the same as the NN for the noiseless case, 3) the NN can do well even with a very small number of training samples, 4) the NN has a good degree of fault tolerance, and 5) the Zernike moment based features possess strong class separability power and are more powerful than moment invariants.

This paper presents a new model-based approach for texture classification which is rotation invariant, i.e., the recognition accuracy is not affected if the orientation of the test texture is different from the orientation of the training samples. The method uses three statistical features, two of which are obtained from a new parametric model of the image called a "circular symmetric autoregressive model." Two of the proposed features have physical interpretation in terms of the roughness and directionality of the texture. The results of several classification experiments on differently oriented samples of natural textures including both microtextures and macrotextures are presented.

In this paper a new set of rotation invariant features for image recognition is introduced . The features are the magnitudes of a set of orthogonal complex moments of the image called Zernike moments . These features could easily be constructed to an arbitrary high order . Taking advantage of the orthogonality property, a systematic feature selection method for choosing an appropriate number of the Zernike features is developed . It is based on computing a measure of the information content differences of features of different classes . The performance of the method is experimentally tested on a 26-class and a I0-class data set involving differently oriented binary images . The first set consists of 624 images of all English characters . The second one is an extensive set of numeric handprinted characters involving 16 .550 samples . Using a nearest neighbor classifier 997E and 84c,% classification accuracies are obtained respectively .

This paper presents a new feature extraction method for classifying a texture image into one of the I possible classes C ;, i= I, . . ., l. It is assumed that the given Mx M image characterized by a set of intensity levels, {y(si, s2), 0 < s] , s1 <_ M-1 }, is a realization of an underlying random field model, known as the Simultaneous Autoregressive Model (SAR) . This model is characterized by a set of parameters 0 whose probability density function p ;(®), depends on the class to which the image belongs . First it is shown that the maximum likelihood estimate (M .L .E .) O •, of 0 is an appropriate feature vector for classification purposes . The optimum Bayes classifier which minimizes the average probability of classification error, is then designed using 0" . Finally the efficiency of the feature vector is demonstrated through experimental results obtained with some natural texture data and a simpler quadratic mean classifier .

This paper describes a totally automatic non-parametric clustering algorithm and its application to unsupervised image segmentation. The clusters are found by mode analysis of the multidimensional histogram of the considered vectors through a non-iterative peak-climbing approach. Systematic methods for automatic selection of an appropriate histogram cell size are developed and discussed. The algorithm is easily parallelizable and is simulated on a SEQUENT parallel computer. Image segmentation is performed by clustering features extracted from small local areas of the image. Segmentation of textured, color, and gray-level images are considered. Eight-dimensional random field model based features, three-dimensional RGB components, and one-dimensional gray levels are utilized for these three types of images respectively. For texture segmentation, an image plane cluster validity procedure based on region growing of the mapped back clusters in the feature space is developed. Most of the phases are also parallelized resulting in almost linear speed ups. Quite satisfactory results are obtained in all cases.

trolled at different operating conditions using a direct torque control (DTC) technique combined with a new switching pattem producing low harmonics. The operating flux of the motor is chosen optimally for both loss minimization and good dynamic response. Since speed estimation is sensitive to rotor resistance variations, the rotor resistance value is calculated and modified in real-time continuously. Simulation and experimental results show that the proposed DTC is able to follow the reference speed (which may be only input reference of the system) with a reasonable dynamic and relatively low error.

One of the problems in increasing reliability in the manufacture of integrated circuit devices is inspection of the bond pads and the bonds connecting the bond pads to the lead fingers of the device. The continuing increase in packing density of VLSI circuits requires that the inspection process be completely automated. Here, we present methods for visual inspection of bond pads and bonds, which are intended to automatically extract parameters of significance in determining their quality, from two-dimensional images taken from the top of the IC wafer. 01484411/93$03.00 0 1993 IEEE

A new method for design of a fuzzy-rule-based classifier using genetic algorithms (GAs) is discussed. The optimal parameters of the fuzzy classifier including fuzzy membership functions and the size and structure of fuzzy rules are extracted from the training data using GAs. This is done by introducing new representation schemes for fuzzy membership functions and fuzzy rules. An effectiveness measure for fuzzy rules is developed that allows for systematic addition or deletion of rules during the GA optimization process. A clustering method is utilized for generating new rules to be added when additions are required. The performance of the classifier is tested on two real-world databases (Iris and Wine) and a simulated Gaussian database. The results indicate that highly accurate classifiers could be designed with relatively few fuzzy rules. The performance is also compared to other fuzzy classifiers tested on the same databases.

This work considers the problem of estimating parameters of multispectral random field (RF) image models using maximum likelihood (ML) methods. For images with an assumed Gaussian distribution, analytical results are developed for multispectral simultaneous autoregressive (MSAR) and Markov random field (MMRF) models which lead to practical procedures for calculating ML estimates. Although previous work has provided least squares methods for parameter estimation, the superiority of the ML method is evidenced by experimental results provided in this work. The effectiveness of multispectral RF models using ML estimates in modeling color texture images is also demonstrated.

This paper presents a novel neural network (NN) approach to the problem of stereopsis. The correspondence problem (finding the correct matches between pixels of the epipolar lines of the stereo pair from amongst all the possible matches) is posed as a noniterative many-to-one mapping. Two multilayer feed-forward NN's are utilized to learn and code this nonlinear and complex mapping using the back-propagation learning rule and a training set. The first NN is a conventional fully connected net while the second one is a sparsely connected NN with a fixed number of hidden layer nodes. Three variations of the sparsely connected NN are considered. The important aspect of this technique is that none of the typical constraints such as uniqueness and continuity are explicitly imposed. All the applicable constraints are learned and internally coded by the NN's enabling them to be more flexible and more accurate than the existing methods. The approach is successfully tested on several random-dot stereograms. It is shown that the nets can generalize their learned mappings to cases outside their training sets and to noisy images. Advantages over the Marr-Poggio algorithm are discussed and it is shown that the NN's performances are superior.

Measurements of high-speed network traffic have shown that traffic data exhibits a high degree of self-similarity. Traditional traffic models such as AR and ARMA are not able to capture this long-range-dependence making them ineffective for the traffic prediction task. In this paper, we apply the fractional ARIMA (F-ARIMA) model to predict one-step-ahead traffic value at different time scales. F-ARIMA has the ability to capture both the short-and long-range dependent characteristics of the underlying data. We present a simplified adaptive prediction scheme to reduce the F-ARIMA computational complexity. The performance of the proposed F-ARIMA prediction model is tested on four different types of traffic data: MPEG and JPEG video, Ethernet and Internet. We also apply the F-ARIMA prediction model to a dynamic bandwidth allocation scheme. The results show that the performance of F-ARIMA outperforms the AR models. They also show that the prediction performance depends on the traffic nature and the time scale.

The paper describes a fuzzy logic controller which sets parameters for the wire bonding process, specifically controlling bonded ball size and shear strength density. The design of the fuzzy engine and the development of rules based on operator experience are discussed. The controller has been tested on actual devices; results show improved process control

This paper describes a color texture-based image segmentation system. The color texture information is obtained via modeling with the Multispectral Simultaneous Auto Regressive (MSAR) random field model. The general color content characterized by ratios of sample color means is also used. The image is segmented into regions of uniform color texture using an unsupervised histogram clustering approach that utilizes the combination of MSAR and color features. The performance of the system is tested on two databases containing synthetic mosaics of natural textures and natural scenes, respectively.

A maritime automatic target recognition system is developed that performs ship classification using one-dimensional high resolution range profiles. Five physically based features are defined and are extracted from both VV and HH polarizations resulting in a 10-dimensional feature vector. A nonlinear classifier combination approach involving a neural network combiner along with three individual classifiers (Bayes, nearest neighbor, and neural network) is proposed. A decision confidence measure based on the classifier discriminants is developed using a nonparametric estimation approach. The confidence measure enables the system to reject samples that have a low decision confidence. The performance of the proposed neural network based combination is compared with individual classifiers and a number of other combination rules. The results show that this combination can provide high recognition accuracy along with a high probability of declaration. The performance in the presence of samples from not-before-seen classes is also investigated. A new nearest neighbor confidence thresholding approach is developed to aid the neural network combiner in rejecting these samples.

This paper describes a novel color texture-based image retrieval system for the query of an image database to find similar images to a target image. The retrieval process involves segmenting the image into regions of uniform color texture using an unsupervised histogram clustering approach that utilizes the combination of Multispectral Simultaneous Auto Regressive (MSAR) and color features. The color texture content, location, area and shape of the segmented regions are used to develop similarity measures describing the closeness of a query image to database images. These attributes are derived from the maximum fitting square and best fillting ellipse to each of the segmented regions. The proposed similarity measure combines all these attributes to rank the closeness of the images. The performance of the system is tested on two databases containing synthetic mosaics of natural textures and natural scenes, respectively.

A novel approach utilized in video database objectbased queries is proposed. This new method segments an example video sequence into real world objects using a combination of color image segmentation techniques along with MPEG-1/2 motion vectors. First, the initial frame in the sequence undergoes a color/texture segmentation algorithm that divides the frame into homogenous regions of color and texture. Features extracted from the co-occurrence matrix provide the texture measurement utilized by this system. The MPEG-1/2 motion vectors are then exploited in predicting the objects of the next frame based on the initial frame's segmentation. A robust technique for the identification of new objects as they enter the scene is proposed along with a measurement pertaining to the object segmentation accuracy is introduced. This new method has a host of visual data mining applications including image database retrieval and Internet image searches. This new algorithm is tested on various image sequences with segmentation results included.