Richard Geoffrey White

Optimum texture analysis of SAR images

This paper demonstrates the importance of texture in SAR image interpretation. Initilly, a simulated annealing technique for despeckling SAR images is outlined. This yield an optimum estimate of the underlying image cross-section. On attempting to apply the method to distinguishing virgin forest from clearings in the Amazon rain forest, we find that the information is not carried in the intensity itself but in the image texture. We discuss the choice of texture estimator that conveys the maximum amount of information and apply the simulated annealing technique to this optimum texture image. We demonstrate that the two categories (forest/clearing) are now almost completely separated with a low misclassification rate. Finally, we examine the sensitivity of the method to the SAR resolution and show that the texture would generally not be visible in an ERS1 image at the conventional incidence angle whereas it could just be with the maximum available incidence angle.

Simulated annealing algorithm for radar cross-section estimation and segmentation

We present here an algorithm which performs radar cross-section estimation by using techniques based on simulated annealing. Standard simulated annealing approaches to image restoration attempt to categorize each image element as belonging to one of a small number of predefined image states or values. This is restrictive for tasks such as radar cross-section estimation and we present here an algorithm which is capable of producing a real-valued output. This is achieved by introducing an edge detection stage into the simulated annealing process. The action of the annealing algorithm may be viewed as a filter which adapts to local image structure. We present results which demonstrate this behavior and in so doing allow us to estimate the residual noise levels we might expect.

Radar cross-section estimation of SAR images

We present an algorithm that is able to smooth out the speckle from many SAR images and which does not suffer from the drawbacks of multilooking. The algorithm is able to preserve the detail and resolution of the original image while producing a smooth, real-valued output. In many cases the quality of the smoothed image is sufficiently high that it may be used with standard optical post-processing algorithms. We use a global optimization method (simulated annealing) and single point gamma statistics to find the MAP solution for the radar cross- section. However, this method may also be regarded as an ideal adaptive filter that is both computationally efficient and highly parallelizable. Results are presented for airborne, ERS-1 and multi-temporal SAR images.

Simulated annealing algorithm for SAR and MTI image cross section estimation

In this paper we first review an algorithm which performs radar cross- section estimation by using techniques based on simulated annealing. Standard simulated annealing approaches to image restoration attempt to categorize each image element as belonging to one of a small number of predefined image states or values. This is restrictive for tasks such as radar cross-section estimation and we present here an algorithm which is capable of producing a real-valued output. This is achieved by introducing an edge detection stage into the simulated annealing process. This original cross-section estimation is based on a flat region model. This is extended here to include linear sloping regions. Results using this new model are given and a qualitative comparison drawn with the original flat region model approach.

Optimum edge detection in SAR

In this paper we derive the maximum likelihood (ML) criterion for splitting (or merging) two regions of single-look SAR imagery as a function of the mean intensity. Two distinct optimization criteria can be postulated: (1) maximizing the total probability of detecting an edge within a window; and (2) maximizing the accuracy with which the edge position can be determined. Initially we derive the ML solution for the first criterion and demonstrate its superiority over an approach based on the Student t test when applied to intensity segmentation. Next we discuss the ML solution for determining the edge position. Finally, we propose a two-stage edge detection scheme offering near optimum edge detection and position estimation.

Cross-section estimation by simulated annealing

Presents an algorithm which performs radar cross-section estimation by using techniques based on simulated annealing. Standard simulated annealing approaches to image restoration attempt to categorise each image element as belonging to one of a small number of predefined image states or values. This is restrictive for tasks such as radar cross-section estimation and the authors present an algorithm which is capable of producing a real-valued output. This is achieved by introducing an edge detection stage into the simulated annealing process. The action of the annealing algorithm may be viewed as a filter which adapts to local image structure. The authors present results which demonstrate this behaviour and in so doing allow them to estimate the residual noise levels that might be expected.<<ETX>>

Real—time SAR change—detection using neural networks

This paper describes the techniques evolved at RSRE for the production of undistorted, focused synthetic aperture radar (SAR) images, target detection using a neural network method and the automatic detection of changes between pairs of SAR images. All these processes are achievable in a single pipelined process operating on an input data rate in excess of 10 Mbytes/second.© (1990) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Smoothing SAR images with neural networks

Describe an approach, to the removal of radar image noise, based on the use of neural networks. A neural network factorisation scheme, based on the use of vector quantisers, allows the authors to produce a more effective solution than that which is possible with a single network. The factorised neural network is currently trained to learn the smoothing behaviour of a noise removal algorithm. The success of the approach demonstrates the potential for this technique and opens the way for its use in learning a true noise smoothing mapping based on the comparison of single and multi look radar data.<<ETX>>

Design of a real-time high-quality SAR processor

Research at the DRA, Malvern, has resulted in a series of algorithms which are capable of yielding focused, undistorted SAR imagery. Unfortunately these can only be implemented in a fraction of a percent of real-time on a standard work-station. In parallel with the algorithm development, therefore, has been research into a real-time implementation on a parallel computer (the Meiko CS1). This paper outlines the principles behind the software architecture design to achieve the desired speed. Processing functions considered include: initial motion compensation (based on accelerometer data), autofocus with phase correction, final processing and an intensity segmentation stage. Real time processing rates of about 10 MBytes/s are now routinely achieved. We indicate the compromises between processor power, available local memory and communications bandwidth needed to achieve real-time operation. Detailed timings derived from the implementation will be presented together with a discussion of the manner in which this could be varied for different SAR configurations. In parallel with the work on producing real-time high quality imagery has gone a program of research into automated image-understanding techniques. This work is now reaching the stage where reliable algorithms for several basic operations, including segmentation and change detections, exist in a form capable of processing continuous imagery at real time or near real-time rates. Provision has been made for the inclusion of these algorithms as postprocessing stages in the real-time SAR processor.

Comparison of neural networks and classical texture analysis

In this paper, it is investigated how closely neural networks can approach the optimum classification of radar textures. To this end, a factorization technique is presented which aids convergence to the best possible solution obtainable from the training data. This factorization scheme is designed to be fully general. The specific performances of the factorized networks are studied, in this radar clutter classification problem, when applied to uncorrelated K distributed images. These results are then compared with the maximum likelihood performance and the performances of various intuitive and approximate classification schemes. Furthermore, preliminary network results are presented for the classification of correlated processes and these results are also compared to results obtained using classical techniques.