Stuart W. Perry

Weight assignment for adaptive image restoration by neural networks

This paper presents a scheme for adaptively training the weights, in terms of varying the regularization parameter, in a neural network for the restoration of digital images. The flexibility of neural-network-based image restoration algorithms easily allow the variation of restoration parameters such as blur statistics and regularization value spatially and temporally within the image. This paper focuses on spatial variation of the regularization parameter.We first show that the previously proposed neural-network method based on gradient descent can only find suboptimal solutions, and then introduce a regional processing approach based on local statistics. A method is presented to vary the regularization parameter spatially. This method is applied to a number of images degraded by various levels of noise, and the results are examined. The method is also applied to an image degraded by spatially variant blur. In all cases, the proposed method provides visually satisfactory results in an efficient way.

Aircraft flight parameter estimation using acoustical Lloyd's mirror effect

A model is developed for the acoustical Lloyds mirror effect observed in the output time-frequency distribution of a microphone located near the ground during the transit of a jet aircraft. The feasibility of using this effect for flight parameter estimation is assessed by a simple Cramer-Rao lower bound analysis. The nonlinear least-squares method and the generalized Hough transform method are formulated for flight parameter estimation. The performances of both methods are evaluated and compared using real acoustic data.

Pulse-length-tolerant features and detectors for sector-scan sonar imagery

This paper presents a neural-network-based system to detect small man-made objects in sequences of sector-scan sonar images created using signals of various pulse lengths. The detection of such objects is considered out to ranges of 150 m by using an experimental sector-scan sonar system mounted on a vessel. The sonar system considered in this investigation has three modes of operation to create images over ranges of 200, 400, and 800 m from the vessel using acoustic pulses of a different duration for each mode. After an initial cleaning operation performed by compensating for the motion of the vessel, the imagery is segmented to extract objects for analysis. A set of 31 features extracted from each object is examined. These features consist of basic object size and contrast features, shape moment-based features, moment invariants, and features extracted from the second-order histogram of each object. Optimal sets of 15 features are then selected for each mode and over all modes using sequential forward selection (SFS) and sequential backward selection (SBS). These features are then used to train neural networks to detect man-made objects in each sonar mode. By the addition of a feature describing the sonars mode of operation, a neural network is trained to detect man-made objects in any of the three sonar modes. The multimode detector is shown to perform very well when compared with detectors trained specifically for each sonar mode setting. The proposed detector is also shown to perform well when compared to a number of statistical detectors based on the same set of features. The proposed detector achieves a 92.4% probability of detection at a mean false-alarm rate of 10 per image, averaged over all sonar mode settings.

A recurrent neural network for detecting objects in sequences of sector-scan sonar images

This work presents a system for detecting small man-made objects in sequences of sector-scan images formed using a medium-range sector-scan sonar. The detection of such objects is considered out to ranges of 200 m from the vessel and while the vessel is in motion. This paper extends previous work by making use of temporal information present in the data to improve performance. The system begins by cleaning the imagery, which is done by tracking objects on the sea bed in the imagery and using this information to obtain an improved estimate of the motion of the vessel. Once the vessels motion is accurately known, the imagery is cleaned by temporally averaging the images after motion compensation. The detector consists of two stages. After the first detection stage has identified possible objects of interest, a bank of Kalman filters is used to track objects in the imagery and to supply sequences of feature vectors to the final detection stage. A recurrent neural network is used for the final detection stage. The feedback loops within the recurrent network allow the incorporation of temporal information into the detection process. The performance of the proposed system is shown to exceed the performances of other models for the final detection stage, including nonrecurrent networks that make use of temporal information supplied in the form of temporal feature vectors. The proposed detection system attains a probability of detection of 77.0% at a mean false-alarm rate of 0.4 per image.

Perception based adaptive image restoration

This paper presents an image restoration technique which uses a cost function based on a novel image error measure. The cost function presented here takes into account local statistical information of the image when performing restoration. It is shown that this technique compares favourably with other techniques, especially when applied to colour images.

Neural vision system and applications in image processing and analysis

We present a computer vision system based on an integrated neural network architecture. In the low level vision subsystem, a network of networks-a biologically inspired network is used to recursively perform filtering, segmentation and edge detection; in the intermediate level and the high level, hierarchically structured arrays of self-organizing tree maps-extension of the popular self-organizing map are utilized to carry out image/feature analysis. The system has been applied to solve a number of real world problems. Some interesting and encouraging results are reported.

A partitioned modified Hopfield neural network algorithm for real-time image restoration

Abstract In this paper, two image partitioning schemes are examined. The first scheme examined avoids boundary conflicts by the use of four restoration phases. The second scheme examined requires a degree of synchronization of the processors restoring adjacent regions. Both schemes avoid conflicting boundary conditions by taking into account the local image formation properties. Without any loss of processing speed, or increase in the number of processors required to restore the image, synchronizing conditions are not required in the four-phase scheme to restore the image accurately, however can be used to maximize restoration efficiency. An improved modified Hopfield neural network-based algorithm is developed to be especially applicable to the problems of real-time image processing based on the described partitioning schemes. The proposed algorithm extends the concepts involved with previous algorithms to enable faster image processing and a greater scope for using the inherent parallelism of the neural network approach to image processing. The simulation in this investigation shows that the new algorithm is able to maximize the efficiency of the described partitioning methods. This paper also presents an example of an application of the proposed algorithm to restore images degraded by motion blur.

Detection of small man-made objects in sector scan imagery using neural networks

This paper presents a neural network based system to detect small man-made objects in sequences of sector scan sonar images. The detection of such objects is considered out to ranges of 150 metres using a forward-looking sonar system mounted on a vessel. After an initial cleaning operation performed by compensating for the motion of the vessel, the imagery was segmented to extract objects for analysis. A set of 31 features extracted from each object was examined. These features consisted of basic object size and contrast features, shape moment-based features, moment invariants, and features extracted from the second-order histogram of each object. The best set of 15 features was then selected using sequential forward selection and sequential backward selection. These features were then used to train a neural network to detect man-made objects in the image sequences. The detector achieved a 97% accuracy at a mean false positive rate of 9 per frame.

An image processing approach for aircraft flight parameter estimation using the acoustical Lloyd's mirror effect

A time-frequency analysis of the output of an acoustic sensor located above the ground during the transit of an aircraft shows an interference (or fringe) pattern on the time-frequency plane. This interference pattern, referred to as the Lloyds mirror effect, is caused by the temporal variations of the constructive/destructive interference frequencies of the direct and ground-reflected aircraft sound fields at the sensor. A model has been developed to describe the temporal variations of the destructive-interference frequencies for an aircraft in level flight over a hard ground. This paper describes two methods to estimate the aircraft flight parameters based on this model. In both methods, the time-frequency distribution of the sensor output is treated as an image. This image is pre-processed to enhance the destructive-interference pattern and then the flight parameters are extracted from the resultant image by optimising a cost function. The effectiveness of the methods is verified using real acoustic data.

Restoration of images degraded by space-variant distortion using a neural network

This paper introduces a neural network algorithm to the restoration of images suffering a known form of space-variant distortion. Using multiple weighting matrices to represent space-variance, the algorithm provides high quality restorations. The algorithm will also be shown to be very computationally inexpensive due to the ability to minimise the neural networks energy in the most efficient manner.