Showing papers on "Contextual image classification published in 1989"

Multivariate Tissue Classification of MRI Images for 3-D Volume Reconstruction - A Statistical Approach

Abstract: One of the major problems in 3-D volume reconstruction from magnetic resonance imaging (MRI) is the difficulty in automating the classification of soft tissues. Because of the complicated soft tissue structures revealed by MRI, it is not easy to segment the images with simple algorithms. MRI can obtain multiple images from the same anatomical section with different pulse sequences, with each image having different response characteristics for each soft tissue. Using the gray level distributions of soft tissues, we have developed two statistical classifiers that utilize the image context information based on the Markov Random Field (MRF) image model. One of the classifiers classifies each voxel to a specific tissue type and the other estimates the partial volume of each tissue within each voxel. Since the voxel sizes of tomographic images are finite and the measurements from tissue boundaries represent the mixture of multiple tissue types, it is preferable that the classifier should not classify each voxel in all-or-none fashion; rather, it should be able to tell the percentage volume of each class in each voxel for the better visualization of the prepared 3-D dataset. The paper presents the theoretical basis of the algorithms and experimental evaluation results of the classifiers in terms of classification accuracy, as compared to the conventional maximum likelihood classifier.

Feature Utility In Polarimetric Radar Image Classificatiion

Abstract: The information content in polarimetric SAR images is examined, and the polarimetric image variables containing the information that is important to the classification of terrain features in the images are determined. It is concluded that accurate classification can be done when just over half of the image variables are retained. A reduction in image data dimensionality gives storage savings, and can lead to the improvement of classifier performance. In addition, it is shown that a simplified radar system with only phase-calibrated CO-POL or SINGLE TX channels can give classification performance which approaches that of a fully polarimetric radar.

Terrain Classification Using Texture For The ALV

Abstract: Off-road navigation is a very demanding visual task in which texture can play an important role. Travel on a smooth road or path can be done with greater speed and safety in general than on rough natural terrain. In addition, recognition of off-road terrain types can aid in finding the fastest and safest route through a given area. Implementations of two texture methods for identifying certain terrain features in video imagery are briefly discussed. The first method uses edge and morphological filters to identify roadways from off-road. The second method uses a neural net to identify several terrain types based on color, directional texture, global variance and location in the image. Plans to integrate the terrain labeled image produced by the latter method into the ALV's perception system are also discussed.

Classification techniques for digital map compression

Abstract: A comparison is made of the performance of various image classification techniques as applied to color cartographic maps is compared. The maps have a lot of graininess due to imperfections in the printing process, which decreases the efficiency of compression techniques. The color maps are classified using the K-means clustering algorithm and vector quantization (VQ), with neighborhood classification to improve the visual quality and compression ratio. The classification is performed in various image representation schemes. The performance of the classifier is evaluated on the basis of the visual quality of the classified image, the time required to classify the image, and compression achieved on the classified image. In terms of computation times, K-means exhibits a clear lead over the VQ classification scheme. However, the VQ classifier converges in fewer iterations than the K-means algorithm. The algorithms eliminated almost all misclassified pixels that were present in the image. The K-means algorithm with neighborhood classification, however, resulted in the filling in of one of the letters and a deterioration in the quality of the lines. >

One-class generalization in second-order backpropagation networks for image classification

Abstract: In an earlier paper, we reported that it is possible to train a first-order multi-layer feedforward network with backpropagation to classify raw 8-bit images of vehicles. We concluded that a linear feedforward network is capable of within-class generalization when trained with perspective views taken every 10{degree}, but it is incapable of one-class generalization. This paper describes the results of a set of experiments to train a feedforward network with second-order inputs to perform one-class classification on image data. We compare the results of the first-order network and the second-order network and show that the second order network is better able to generalize as a one-class classifier. 7 refs., 6 figs.

Pattern Recognition Using Photon-Limited Images

Abstract: The spatial coordinates of detected photoevents and the number of detected photoevents in a given area convey information about the classical irradiance of the input scene. In this paper the effectiveness of photon-counting techniques for image recognition is discussed. A correlation signal is obtained by cross correlating a photon-limited input scene with a classical intensity reference function stored in computer memory. Laboratory experiments involving matched filtering, rotation- and scale-invariant image recognition, and image classification are reported. For many images it is found that only a sparse sampling of the input is required to obtain accurate recognition decisions, and the digital processing of the data is extremely efficient. Using available photon-counting detection systems, the total time required to detect, process, and make a recognition decision is typically on the order of tens of milliseconds. This work has obvious applications in night vision, but it is also relevant to areas such as process control, radiological, and nuclear imaging, spectroscopy, robot vision, and vehicle guidance.

Two-Dimensional Discrete Gaussian Markov Random Field Models for Image Processing and Analysis

Abstract: This paper is concerned with a systematic exposition of the usefulness of two-dimensional (2-D) discrete Gaussian Markov random field (GMRF) models for image processing and analysis applications. Specifically, we discuss the following topics; notion of Markovianity on a plane, statistical inference in GMRF models; and their applications in several image related problems such as, image synthesis, texture classification, segmentation and image restoration.

Range Image Analysis Via Quadtree And Pyramid Structure Based On Surface Curvature

Abstract: This paper presents a method which can expedite, while maintaining the accuracy, range image analysis such as range image segmentation, classification, and location of the region of interest in the range image, thereby making 3-D vision techniques based on range images more useful in manipulating robots accurately in real time. The proposed method incorporates quadtree and pyramid structure in order to quickly analyze range images. In order to make range image analysis independent of the viewing directions, surface curvatures, which are visible-invariant surface characteristics, are exploited. Specifically, we will discuss the following topics: (1) problems of using the surface curvatures for the range image analysis in the presence of noise; (2) generation of the range image pyramid; (3) reliable range image segmentation and classification via split-and-merge using the planarity test and the surface curvatures; (4) incorporation of the quadtree and the pyramid structure to speed up the projection process.

Per-point and per-field contextual classification of multipolarization and multiple incidence angle aircraft L-band radar data

Abstract: Multipolarized aircraft L-band radar data are classified using two different image classification algorithms: (1) a per-point classifier, and (2) a contextual, or per-field, classifier. Due to the distinct variations in radar backscatter as a function of incidence angle, the data are stratified into three incidence-angle groupings, and training and test data are defined for each stratum. A low-pass digital mean filter with varied window size (i.e., 3x3, 5x5, and 7x7 pixels) is applied to the data prior to the classification. A predominately forested area in northern Florida was the study site. The results obtained by using these image classifiers are then presented and discussed.

Multi-Pass Multi-Resolution Image Interpretation

Abstract: Previous work in automatic photointerpretation has performed feature extraction and identification as one-pass sequential processes, at a single global level of detail. However, human photointerpreters perform these actions simultaneously while focusing on local areas of interest or uncertainty, at an appropriate level of detail. This paper describes our system for a piece-wise approximation to human-like photointerpretation, called Multi-Pass Multi-Resolution (MPR) image interpretation. MPR uses multiple passes to approximate simultaneity, and multiple resolutions together with recursive segmentation to achieve varying levels of local detail.

Pattern Recognition Based On Circular Scanning Of Images

Abstract: Circular scanning of images enables simple feature extraction basically insensitive to object orientation and position. Such is a histogram representing concentric luminous intensity of the image. The histogram shows luminous intensity of concentrinc rings having center in object's centroid. However, features extracted from the image in such a way do not uniquely describe an object. This can cause certain difficulties in pattern recognition stage of image analysis. Latin capitals have been chosen as the objects to be recognized. Two approaches have been analysed. In the deterministic approach pattern recognition is based on matching of the features of an object with those from the feature space. Satisfactory matching results in recognition of the particular capital. Small changes in shapes of the capitals to be recognized can show applicability of the described method in practice. The other approach is based on simple form of adaptive recognition. The result of recognition changes the feature space in order to increase reliability of further recognition. Averaging the data from feature space with the features extracted from the recognized image increase the reliability of further recognition. Experimental results were obtained in problem of recognition of the six capitals given in two different fonts and scales. Images of the capitals were obtained by means of binary semiconductor camera. Image processing, feature extraction and pattern recognition were performed on IBM PC AT.

Multi-temporal classification of remote sensed data in image processors with a pipeline architecture

Abstract: Hierarchical classifiers are discussed and a new implementation in digital image processors with a pipeline architecture is presented. This implementation makes optimal use of the different layers of look up tables that generally exist in all machines with this architecture. Classification can be considered to occur at almost real time, as far as the unit structure of the tree associated to the classifier is concerned. As the hierarchical classifier makes optimal use of all the information available, fast decision rules can be applied in most of the nodes, thus reducing considerably the overall computational burden. An example is provided in the context of classification of multi-temporal remote sensed data. To apply the technique, the evaluation of the features must have been done previously.

Scene Matching: Photon By Photon

Abstract: The spatial coordinates of detected photoevents in a given area convey information about the classical irradiance of the input scene In this paper the effectiveness of photon-counting techniques for image recognition is discussed A correlation signal is obtained by cross correlating a photon-limited input scene with a classical intensity reference function stored in computer memory Laboratory experiments involving matched filtering, rotation-invariant image recognition, and image classification are reported For many images it is found that only a sparse sampling of the input is required to obtain accurate recognition decisions, and the digital processing of the data is extremely efficient Using available photon-counting detection systems, the total time required to detect, process, and make a recognition decision is typically on the order of tens of milliseconds This work has obvious application in night vision, but it is also relevant to areas such as robot vision, vehicle guidance, radiological and nuclear imaging, and recognition of spectral signatures

Prospects for Classifying Complex Imagery Using a Self-Organizing Neural Network

Abstract: : The objective of this study is to evaluate the performance of Fukushima's Neocognitron model when it is applied to complex imagery. This system could discriminate between simple alphabetical characters represented in fields of 16 by 16 pixels, and that shift invariance can be achieved through a proper choice of design parameters. This work describes results for expanded Neocognitron architectures operating on complex images of 128 by 128 pixels. These neural network systems were simulated on a VAX-8600 minicomputer. Wire frame models of three different vehicles were used to test the properties which Fukushima had demonstrated. The expanded Neocognitron systems were able to classify these objects and to identify their critical features. After training, each object was placed at different positions in the plane, and the Neocognitron's shift invariance property was tested. With complex (128 X 128) imagery, it was difficult to achieve proper classification and maintain shift invariance using only a few levels. In another experiment, the Neocognitron trained on polar transforms of objects in the training set. Objects in the training set were rotated, and polar transforms of the rotated images were submitted as input. In this manner, the Neocognitron's shift invariance was exploited to recognize rotated imagery. These investigations gave insight into the role of various model parameters and their proper values, as well as demonstrating the model's applicability to complex images. Keywords: Neural network; Shift invariance; Rotation invariance; Image classification; Unsupervised learning; Multilayer architecture; Parallel computing.

The Interval Probability Density As A Tool For Image Classification

Abstract: This work describes a statistical function , M(A) , obtained by averaging cross products of the time-spatial interval probability density.In some cases, as for example deterministic low light level signals, M(Li) contains the same information as the autocorrelation function with an important improvement of the SNR.T measurement of the MC can be applied in image classification in those cases which we have the value of g of the reference images.

Filter Synthesis Using Neural Networks

Abstract: We are investigating an opto-electronic implementation of a trainable pattern classification system based on a feed-forward neural network model. An architecture with two layers of interconnections is used to transform a large amount of scene information to a small feature space that is, in turn, transformed into classification data. By using two layers of interconnections the number of large inner products that must be calculated may be significantly reduced. Simulations have been performed on a digital computer that demonstrate the performance of a system for the rotation-invariant classification of printed characters. A possible optical implementation is outlined.

Clustering and Classification for Chinese Character Recognition

Abstract: The paper evaluates the applicability and results of several clustering and classification algorithms for optical Chinese character recognition. Emphases are laid on k-means clustering algorithms, Neural Nets classification, and Hidden Markov Model matching scheme. Some experimental results of the algorithms are also presented.