Showing papers on "Contextual image classification published in 1987"

Classification of Remotely Sensed Images

Abstract: Remotely sensed images are an indispensable tool to researchers in disciplines such as cartography, forestry and geology in which mapping by satellite or aircraft is often a far more acurate, efficient and cost-effective method than conventional survey. Most users of remote sensing data are now familiar with using photographic images prepared to standard prescriptions by others to relate colours or grey tones to what actually exists on the ground. This book is aimed at the discipline-oriented user who wishes to move away from these traditional photographic interpretation methods towards interactive analysis systems. Such systems permit the classification of the digital data in ways that enable locations and quantitative information on specific themes to be extracted and portrayed. The empasis is upon the integration of these new techniques into existing discipline skills: the user is not expected to become a 'computer operator'. The book covers the fundamental theory of image classification, and illustrates it with practical examples in the form of a structured outline of a total research project. It is therefore an essential reference manual to 'sit at the elbow' of the user working with an image processing system for remotely sensed data.

Multispectral magnetic resonance image analysis.

Abstract: Multiecho magnetic resonance (MR) scanning produces tomographic images with approximately equal morphologic information but varying gray scales at the same anatomic level. Multispectral image classification techniques, originally developed for satellite imaging, have recently been applied to MR tissue characterization. Statistical assessment of multispectral tissue classification techniques has been used to select the most promising of several alternative methods. MR examinations of the head and body, obtained with a 0.35, 0.5, or 1.5T imager, comprised data sets with at least two pulse sequences yielding three images at each anatomical level: (1) TR = 0.3 sec, TE = 30 msec, (2) TR = 1.5, TE = 30, (3) TR = 1.5, TE = 120. Normal and pathological images have been analyzed using multispectral analysis and image classification. MR image data are first subjected to radiometric and geometric corrections to reduce error resulting from (1) instrumental variations in data acquisition, (2) image noise, and (3) misregistration. Training regions of interest (ROI) are outlined in areas of normal (gray and white matter, CSF) and pathological tissue. Statistics are extracted from these ROIs and classification maps generated using table lookup, minimum distance to means, maximum likelihood, and cluster analysis. These synthetic maps are then compared pixel by pixel with manually prepared classification maps of the same MR images. Using these methods, the authors have found that: (1) both supervised and unsupervised classification techniques yielded theme maps (class maps) which demonstrated tissue characteristic signatures and (2) tissue classification errors found in computer-generated theme maps were due to subtle gray scale changes present in the original MR data sets arising from radiometric inhomogeneity and spatial nonuniformity.

Multispectral analysis of MR images of the breast.

Abstract: Preliminary investigations were conducted into the potential of magnetic resonance (MR) images for tissue classification of the breast on the basis of relative signal intensity. Multispectral techniques originally developed by the National Aeronautics and Space Administration for satellite image analysis were used in sequence selection, image data correction, image standardization, and image interpretation. Numerous sequence combinations with varying repetition times (TR) and echo times (TE) were considered, and a triplet was selected consisting of long TR/long TE, short TR/short TE, and an opposed phase sequence with intermediate TR and TE. Correction to remove system-imposed intensity inhomogeneities was required for all images. Image standardization based on fat and pectoral muscle signals was necessary for intercase comparisons. Multispectral images obtained based on this analysis suggest the feasibility of intensity-based image classification.

Monitoring rice areas using LANDSAT MSS data

Abstract: A specialized system to monitor the rice growing areas of New South Wales has been developed. The monitor has been run for 2 years with accurate and consistent results. The monitor system starts with a land-cover mask. Each image is classified and the result used to update the mask using agronomic criteria. The criteria can, and often will, be changed between the different phonological stages of the crop. The system, with incremental revision of a land-cover mask, is quite different in approach to that adopted in multi-temporal image classification as utilized in LACIE and AgRISTARS.

Bayesian classification of surface-based ice-radar images

Abstract: This paper deals with the application of the Bayes classification procedure to discriminate types of sea ice based on images obtained from surface-based marine radars. The data sets were digitized images obtained from a dual-polarized Ku -band radar (16 GHz) and a like-polarized S -band radar (3 GHz) at a site located on the northern tip of Baffin Island, Canada. The images were first range-compensated, and statistical properties of different ice types were then determined. The observed histograms for different ice types were approximated by continuous density functions. The images were classified by maximizing the a posteriori probabilities obtained from Bayes's rule. The results reported herein suggest that there is sufficient information in the reflectivity to classify the different forms of ice using decision-theoretic pattern recognition techniques.

Linear Maps with Point Rules: Applications to Pattern Classification and Associative Memory

Abstract: Generalisations of linear discriminant functions are introduced to tackle problems in pattern classification, and associative memory. The concept of a point rule is defined, and compositions of global linear maps with point rules are incorporated in two distinct structural forms—feedforward and feedback—to increase classification flexibility at low increased complexity. Three performance measures are utilised, and measures of consistency established. Feedforward pattern classification systems based on multi-channel machines are introduced. The concept of independent channels is defined and used to generate independent features. The statistics of multi-channel classifiers are characterised, and specific applications of these structures are considered. It is demonstrated that image classification invariant to image rotation and shift is possible using multi-channel machines incorporating a square-law point rule. The general form of rotation invariant classifier is obtained. The existence of optimal solutions is demonstrated, and good sub-optimal systems are introduced, and characterised. Threshold point rules are utilised to generate a class of low-cost binary filters which yield excellent classification performance. Performance degradation is characterised as a function of statistical side-lobe fluctuations, finite system space-bandwidth, and noise. Simplified neural network models are considered as feedback systems utilising a linear map and a threshold point rule. The efficacy of these models is determined for the associative storage and recall of memories. A precise definition of the associative storage capacity of these structures is provided. The capacity of these networks under various algorithms is rigourously derived, and optimal algorithms proposed. The ultimate storage capacity of neural networks is rigourously characterised. Extensions are considered incorporating higher-order networks yielding considerable increases in capacity.

Digital Image Texture Analysis Using Gray Level and Energy Cooccurrence

Abstract: In this paper, a new digital image texture analysis method, i.e., gray level and energy cooccurrence image texture analysis method, is presented. According to the method, the two image texture analysis models are designed. They are 3 X 3 no-zero-sum mask model and gray level-difference cooccurrence matrix model. And their texture parameters are calculated by using these models, separately. The satisfactory test results are acquired for classifying the sensing images.

Knowledge-based classification of landsat thematic mapper digital imagery

Abstract: An innovative approach to identifying land use and land cover categories through computer-assisted analyses of digital satellite remote sensing image data is presented. The technique is based on the subfield of artificial intelligence known as expert systems, commonly referred to as knowledge-based systems. The methodology developed embodies expert image analyst rules and heuristics in a knowledge base which is used to classify regions of a Landsat Thematic Mapper image. Expert knowledge about and image attributes from spectral, spatial, and temporal domains are addressed. The procedures for image and ancillary data preprocessing, knowledge acquisition, knowledge-based image analysis, and traditional image classification are discussed. Both the deterministic knowledge-based image analysis approach and the traditional statistical maximum likelihood approach were applied to multidate Landsat Thematic Mapper digital imagery to derive land use and land cover information. It was found that a knowledge-based approch to the classification of specially-processed, digital remote sensing imagery, coupled with spatial information, produced results superior, in terms of accuracy and visual comprehensibility, to those achieved through conventional per-pixel, supervised classification of multispectral data alone. It was illustrated that the knowledge-based method developed permits the inclusion of heuristics and decision criteria not possible in the strictly numerical, algorithmic approach. Because of the success of this prototype knowledge-based image analysis system, and because of its parallelism with human visual image analysis processes, additional research into this area is recommended and briefly discussed.

Temporal Adaptive Vector Quantization For Image Sequence Coding

Abstract: Vector quantization (Vq) provides us with an appropriate technique to obtain high compression ratios. However, for image sequence coding, vector quantizer has to design temporally stable codebooks of representative vectors and this constraint of temporal stability has to be solved with a minimal computational complexity. We propose a new coding scheme based on visual classification and temporal refreshment procedures which enable to design subjectively optimal codebooks for any kind of real broadcast image sequence. This contribution reports results of simulations about how such a classification may be introduced and evaluated and how vector quantizer may be refreshed. We obtain significantly better reconstruction quality for a given codebook size and investigate several available solutions to implement the codebook refreshment procedures.

Using Contextual Information In Classification Algorithms

Abstract: Some of the most important phases in the processing of digital images are image segmentation and classification. Contextual algorithms make use of information coming from the pictorial content (called spatial or contextual information) of the data in addition to the spectral properties. A 2-dimensional stochastic model is introduced in this paper, which enables us to combine spatial and spectral information in uniform manner. The Bayesian approach is adopted and, according to this, a pixel is assigned to that class which gives the maximum "a pos-teriori" probability conditioned on not only the observed feature vector of this pixel but also of its neighbours. Depending upon the size of the neighbourhoods and the model to compute the joint probabilities, various (formerly known or new) contextual algorithms can be derived. In order to reduce the computational complexity of the algorithms an iterative approximation is proposed and analysed. These suboptimal algorithms give considerably better results (in comparison with non-contextual ones) while time and storage requirements remain acceptable.

An Expert Image Analysis System For Chromosome Analysis Application

Abstract: This paper reports a recent study on applying a knowledge-based system approach as a new attempt to solve the problem of chromosome classification. A theoretical framework of an expert image analysis system is proposed, based on such a study. In this scheme, chromosome classification can be carried out under a hypothesize-and-verify paradigm, by integrating a rule-based component, in which the expertise of chromosome karyotyping is formulated with an existing image analysis system which uses conventional pattern recognition techniques. Results from the existing system can be used to bring in hypotheses, and with the rule-based verification and modification procedures, improvement of the classification performance can be excepted.

Image Processing And Perimetric Sampling For Pattern Recognition

Abstract: This paper examines the classification potential of three techniques based on spiral sampling of gray-scale, noisy images. Image pixels are rearranged into a one-dimensional sequence by selecting samples in a spiral manner starting from the edge of the image and proceeding toward the center. The properties of this sample sequence are examined by Fourier transform and correlation techniques, using images from 26 groups with varying contrast, orientation, and size. The classification ability of features extracted from spiral sequences and their accuracy are investigated.

Contextual Analysis Of Ultrasonic B-Mode Images.

Abstract: In order to reduce the classification errors when labeling texture pixels, the contextual information can be used to reduce the ambiguity attached to pixel classification. A post-processing approach is introduced whereby the context is modeled according to a Markov-Gibbs assumption to describe the local characteritics of a pixel label. Results of some pathological cases are presented.

Pattern Recognition With A Spiral Sampling Technique

Abstract: This paper examines the pattern recognition potential of spiral sampling applied on gray - scale, noisy images. Image pixels are rearranged into a one dimensional sequence by selecting samples in a spiral manner starting from the edge of the image and proceeding toward the center. The properties of this sample sequence are examined by Fourier transform analysis, using images from a number of groups, with varying contrast, orientation and size. The classification ability of features extracted from spiral sequences and their accuracy are investigated.

A Hierarchical Method Of Remotely Sensed Multispectral High Resolution Image Classification

Abstract: One of the simplest classification algorithm which utilizes a linear discriminant function is known as the minimum distance classifier, it is widely used in pattern recognition. But, it encounters the problem of useless dimension compensation when the feature dimensionality is very large. This is the situation when one wants to use textural features as input parameters for classification as it is now possible with the remotely sensed high resolution images (optical or radar). To avoid this problem, we propose a hierarchical classification algorithm.© (1987) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Registration Of Rotated Images

Abstract: The automatic determination of local similarity between two images (image registration) is one of the most fundamental problems of image processing and pattern recognition A two-stage registration algorithm that is reasonably efficient and robust for rotational and translational displacement has been considered to determine relative shift between reference and search images A new method of calculation for zero-order moment on a circular window is described, Based upon this measure, a computationally efficient first stage algorithm is obtained On the second stage, Modified Circular Symmetric Autoregressive Random (MCSAR) model and invariant moments are used to determine a robust feature vector associated with the reference and subsearch images Simulation results on real images are presented to evaluate rotation invariancy of the features and the performance of the proposed algorithm