Abstract
This thesis investigates the process of television standards conversion. That is converting a television signal originated in one standard for display in another standard. A typical example of this process is converting between European television, with 625 lines and 25 frames/second, and American television, with 525 lines and 30 frames/second. Although European/American interconversion is the best known type of standards conversion many other types are becoming necessary or desirable. These other standards conversion processes are between the burgeoning number of standards for high definition television, computer graphics as well as conventional television. The standards conversion process is, essentially, one of resampling a three dimensional sampled signal on a new sampling lattice. In one dimension the analogous process of sample rate changing is well understood. For standards conversion the theory of sample rate changing must be extended to three dimensions. Television standards conversion is much more difficult than sample rate changing an audio signal. This is partly because of the signal is 3 dimensional and partly because the sampling rates are orders of magnitude greater. The most significant problem of standards conversion, however, is the fact that television signals are undersampled spatially and, most significantly, temporally. Undersampling in television signals results in aliasing which confounds the assumptions underlying the theory of sample rate changing. At the start of this work the state of the art in television standards conversion involved interpolation of the signal using a 16 tap, 2 dimensional, finite impulse response filter. The filter coefficients were determined empirically to minimise the picture artifacts caused by the aliasing inherent in the signal. The first purpose of the work described here was to analyse the standards conversion process and develop objective methods of optimising the performance of the existing type of standards converters. It was likely that even optimised standards converters of this type would generate undesirable artifacts in their output pictures. Therefore, the second purpose of the work was to develop improved techniques for standards conversion. To optimise the performance of conventional standards converters it is necessary to analyse the component parts of the television signal chain. Of particular significance for standards conversion are the characteristics of real scenes and the way in which the eye perceives them. The thesis starts with an analysis of the television signal chain. Experimental work confirms previously published results regarding the spectral content of typical scenes and extends them from 2 to 3 dimensions. A model is derived to describe these results and a theoretical justification is given for this model. The frequency response of the human visual system is a important part of the television signal chain and a model is presented for this response, distilled from an extensive review of the literature. A new method is presented for optimising the performance of conventional standards converters using the preceding analysis of the television signal chain. The optimisation technique defines an 'ideal' interpolation filter response which simultaneously minimises picture impairments due to loss of detail and aliasing. For practical standards converters it is necessary to produce an optimal realisable filter which approximates the 'ideal' response. Several new techniques are presented for generating optimum, practical, finite impulse response filters. Although the filter design process is, essentially, performed in the frequency domain some of the filter design methods also allow the inclusion of time domain constraints. There are many ways in which practical television standards conversion hardware can be built. In practice only a very few of the possibilities have been explored. The best way to build a standards converter depends on the type of conversion (eg whether the number of lines is increased or decreased) and the commercially available integrated circuit building blocks. The various ways in which standards converters can be built is examined in a unified and systematic way. This allows the best standards conversion architecture to be selected for a particular application. (Abstract shortened by ProQuest.).