Bailey, Samuel Gerard; (2000) A procedure for designing and assessing the performance of image processing systems. Doctoral thesis (Ph.D), UCL (University College London).

Text A_procedure_for_designing_and_.pdf Download (10MB)

Abstract

The field of image processing (IP) currently lags behind many other fields in science and engineering in the development of techniques for predicting and assessing system performance. This thesis describes a technique for assisting developers and users of IP systems. It presents a methodology for the design and the performance prediction of those systems in different imaging conditions. The thesis surveys various performance analysis techniques which have been developed to analyse IP system performance, namely benchmarking, performance evaluation and performance characterisation. It outlines the differences, as well as the advantages and short-comings of each technique. The thesis then presents a new methodology to guide system designers in gathering the appropriate data about imaging conditions, designing the IP system, and predicting and assessing its performance. The methodology operates by guiding the developer through the following stages: Firstly appropriate parameters are selected to describe the imaging conditions and the final performance metrics. These are narrowed down until only the most important factors remain. The nature of these parameters is then used to determine the best approach to performance analysis, either analytical, empirical, or a combination of the two. An example algorithm is then chosen which could be used to perform the IP task. This algorithm is then modularised, or broken down into its constituent components. These modules are then analysed one by one to determine which imaging parameters affect which module, and what internal quality propagation parameters can be used to measure the effect that the performance of each has on the other modules. Transfer functions are then derived which relate how incoming parameters effect outgoing metrics for each module. Finally, the performance of the different modules is combined, together with a distribution of the operating conditions to produce a final performance measure for the system. The effectiveness of the methodology is demonstrated by applying it to four industrial image-processing systems: visual tracking of batches of steel, automatic identification of batches of steel, lens aberration determination in a transmission electron microscope and fuel drum location for automatic materials handling. In each case an example algorithm is chosen to perform the task, and its performance is predicted under the operating conditions it is likely to encounter. The methodology's applicability to two further tasks is also shown, and conclusions and recommendations for further work are discussed.

Download activity - last month

Export as XMLJSONCSV

Download activity - last 12 months

Export as XMLJSONCSV

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item