GB2243514A

GB2243514A - Generation of interlaced television signals by interpolation

Info

Publication number: GB2243514A
Application number: GB9102814A
Authority: GB
Inventors: Martin Weston
Original assignee: British Broadcasting Corp
Current assignee: British Broadcasting Corp
Priority date: 1990-02-09
Filing date: 1991-02-11
Publication date: 1991-10-30
Anticipated expiration: 2011-02-11
Also published as: GB2243514B; GB9102814D0

Abstract

In order to generate an output video signal of interlaced form from an input signal of the same field rate, weighted contributions from lines of a plurality of fields of the input signal are taken to make each output line. Where the input signal is an interlaced signal at a higher number of lines, the input signal is as a first step converted to a non-interlaced signal having the said higher number of lines per field. The input signal is, as part of the interpolation, filtered with a filter function as illustrated in Figure 7. <IMAGE>

Description

GENERATION OF INTERLACRn TV SIGNALS BY INTERPOLATION 1. Introduction It is often necessary to convert TV signals scanned according to one scanning standard to another. For example the signals derived in the new "High Definition" formats will often need to be included in conventional definition programs, and vice versa.

It is known from our United Kingdom patent application No.

8626066 published under the number 2197152A, that interlaced input signals can be converted to other standards. However, if conversion between two interlaced signals, which is, in fact, the most ctrrn requirement, is to be achieved, it must also be possible to convert to an interlaced output signal standard.

In the case where conversion between two interlaced standards is required, the standard with the lower scanning rate will tend to dominate. So, conversion to an interlaced output standard tends to involve down-conversion to a lower scanning rate (e.g. 1250/50/2 to 625/50/2), whereas conversion from an interlaced input mostly involves upconversion from a lower rate (e.g. 625/50/2 to 1250/50/2).

In accordance with the present invention, there is provided apparatus and a method of generating an output video signal of interlaced form from an input signal of the same field rate, in which the lines of each output field are made by taking weighted contributions from lines of a plurality of fields of the input signal. Where the input video signal is an interlaced signal at a higher number of lines per picture than the output video signal, the method preferably includes as a first step converting the input video signal to a non-interlaced signal having the said higher number of lines per field.A preferred method in accordance with the invention will now be described in detail, by way of example only, with reference to the drawings; in which: Figure 1 shows alias ; Figure 2 shows the effect of filtering on aliassing; Figure 3 shows the variation of visibility with losses as a function of input frequency; Figure 4 shows the visibilities of losses at 0Hz and of the corresponding aliases at 25 Hz; Figure 5 shows the visibilities of losses at 25Hz and of the corresponding aliases at 0Hz; Figure 6 shows the vertical responses of a pre-filter in accordance with the invention at various frequencies; Figure 7 shows the coefficients pattern of a three field interpolator in accordance with the invention; Figure 8 shows the actual response of the interpolator of Figure 7; and Figure 9 shows an optimised coefficient pattern.

2. Interpolation Conversion from one scanning standard to another involves the interpolation of new TV lines lying between the input lines. This is performed by generating each output line from a weighted sum of input lines. The weighting coefficients vary from line to line depending on the relative positions of input and output lines.

An interpolator may be considered as two filters in cascade.

The first fills in the gaps between the input lines, attempting to recreate as accurately as possible the original image. In the frequency domain this filter removes the alias components caused by the input scanning standard. Patent application 8626066 referred to above describes the design of such a filter for interlaced input standards.

The second filter removes any remaining components which would otherwise cause serious aliasing when the signal is subsequently resampled by the output scanning standard. It is the design of this second filter which is the subject of this patent application.

3. Non-Interlaced Outputs 5vamping a signal with a 625-line TV raster repeats the spectrum of the input signal, as shown in figure 1. These repeat spectra, called "aliases", are centred on 625 cycles/picture height (c/ph), and all integer multiples of it.

Most of the highest frequency alias components will be rerrroved (or at least greatly attenuated) by resolution losses in the display device and by the limited resolution of the viewers eyes. But if the input spectrum extends to high enough frequencies, some alias components may extend down to very low frequencies and thus give very visible alias patterning. To prevent this the highest input frequencies should be reproved by a low-pass filter, before sampling with the 625-line raster. The result is shown in figure 2.

The optimum position of the cut-off is a compromise between aliasing and loss of resolution. Lowering the cut-off reduces the aliasing, but increases the loss of resolution. Conversely, raising the cut-off reduces the loss of resolution but allows through more aliasing. So the optimum position depends on the relative visibilities of aliases and resolution losses.

Figure 3 shows how the visibility of losses varies as a function of the input frequency. (This curve is taken from data published in BUDRIKIS Z.L. 1973 'M5del approximations to visual spatio-temporal sine-wave threshold data'. Bell Systems Technical Journal, 1973, 52, pp.1643-1667 for a viewing distance of 6 times picture height). The aliases are frequency shifted versions of the input, so their visibility (as a function of the input frequency which produced them) might be expected to be a frequency shifted version of the visibility of losses, as shown the drawing.

The optimum position of the cut-off is at 312 c/ph, where the two curves cross. This is because moving it higher would include aliasing which is more visible than the corresponding reduction in loss, and moving it lower would give losses which are more visible than extra aliases removed.

The ideal shape of the cut-off is not actually infinitely sharp cut. It can be shown that, to minimise the sum of the squares of the loss and the alias, the ideal filter response R(f) is given by: R(f) -t2(f) / (L2(f) +A2(f)) where L(f) and V(f) are the visibilities of losses and aliases as functions of input frequency f.

In practice, because aliases constitute new unexpected signals, they are usually more noticeable and annoying than losses, so they might perhaps also be scaled by an extra "annoyance factor". This would lower the optimum cut-off frequency.

4. Interlace Scanning the output with an interlaced raster generates a second set of alias components centred on 312.5 c/ph, and with a temporal frequency (flicker) of 25Hz. If it were not for this 25Hz component, the cut-off frequency would need to be lowered to about 156 c/ph.

However, the visibilities of impairments such as aliasing and loss of resolution depend on the temporal frequencies involved. So, because of the 25Hz offset, the optimum position of the cut-off depends on the temporal frequency of the input signal.

Because of interlace, stationary inputs at OHz give aliases at 25Hz. These are less visible than components at OHz. So a higher cut-off frequency may be used, since this gives an increased resolution which is more visible than the corresponding increase in aliasing. This is illustrated by figure 4, which shows the visibilities of losses at OHz and of the corresponding aliases at 25Hz (combined with those centred on 625 c/ph). As before the optimum cut-off is where the two curves cross.

Conversely, moving inputs, with components at 25Hz, give aliases at OHz, which are highly visible. So at 25Hz, a lower cutoff should be used to give a reduction of aliasing, since this is more visible than the corresponding loss of resolution. This is illustrated by figure 5, which shows the visibilities of losses at 25Hz and of the corresponding aliases of OHz. Once again the optimum cut-off is where the two curves cross.

Similarly, the optimum cut-off may be determined for all other temporal frequencies. Figure 6 shows the vertical responses (at 0, 6.25, 12.5, 18.75, & 25Hz) of a pre-filter designed by computer to minimise the sum of the squares of the visibilities of aliasing and losses, when viewing 625/50/2 at a distance of 6 times picture height.

5. Implementation As mentioned in section 2 above, the filter would be combined with an interpolator designed to remove the aliases generated by the input standard. The response of the combined interpolator is the product of the response of the individual filters. Since in practice both responses are fairly sharp-cut, the filter with the lower cut-off tends to dominate, and the filter with the higher cut off can often be ignored. Thus for down-conversion the interpolation is determined mainly by the output standard, whereas for up-conversion it is the input standard which is most important.

Making a filter with a frequency response which depends on temporal frequency requires the use of contributions from more than one input field. Figure 7 shows the coefficients pattern (inpulse response) of a 3-field interpolator, for down-conversion from 1250/50/2 to 625/50/2, with a vertical aperture of 4 output picture lines. This was designed by computer optimisation to minimise the visibilities of the departure of its frequency response from the ideal filter shown in figure 6. The actual response is shown in figure 8.

It is also possible to include correction for losses in the camera and display. Since the interpolator uses contributions from more than one field it is even possible to correct, not only for vertical resolution losses, but also for losses in the temporal direction, such as camera integration and lag. Figure 9 shows an optimised coefficient pattern, for conversion from 1250/50/2 to 625/50/2, assuming typical camera and display losses.

6. Conclusions Conversion to an interlaced standard may be improved by using interpolators which take contributions from more than one input field, even when the two standards have the same field frequency.

Compared to single field interpolators, stationary areas of the picture may be sharper, and moving areas have less visible aliasing.

The interpolation may also correct for the vertical and temporal responses of camera and display.

Although it would seem that, where the interlaced signal is to be generated from a signal having the same field rate, all the information necessary to provide an interlaced signal with good resolution would be found in a single field, we have, surprisingly, found that resolution is improved by taking contributions from more than one field of the input signal.

Claims

1. A method of generating an output video signal of interlaced form from an input signal of the same field rate, in which the lines of each output field are made by taking weighted contributions from lines of a plurality of fields of the input signal.

2. A method according to claim 1, in which the weighted contributions are selected to filter the signal with a filter function which has a lower cut-off frequency in the vertical direction for temporal frequencies of half the field frequency and odd multiples thereof than for temporal frequencies in the region of zero and even multiplies.

3. A method according to claim 2 in which the filter function comprises for each output line taking contributions from a plurality of lines of the corresponding input field and from a plurality of lines of the adjacent input fields, the contributions from certain lines to either side of the corresponding line in the adjacent field being negative and the contributions from lines between the said certain lines being less than the contributions from the associated lines of the current field.

4. A method according to claim 3 in which the filter function is substantially as shown in Figure 7 or Figure 9 of the drawings.

5. A method according to any preceding claim, in which the input video signal is an interlaced signal at a higher flutter of lines per picture than the output video signal, and in which the method includes as a first step converting the input video signal to a noninterlaced signal having the said higher number of lines per field.

6. A method according to any of claims 1 to 4 which operates directly on an interlaced signal.

7. A method according to any preceding claim in which the input video signal is at the 1250/50/2 standard and the output video signal is at the 625/50/2 standard.

8. A method of generating an output video signal of interlaced form from an input video signal of the same field rate, the method being substantially as herein before described with reference to the drawings.

9. Apparatus for generating an output video signal of interlaced form from an input signal of the same field rate, the apparatus comprising means for taking weighted contributions fram lines of a plurality of fields of the input signalto make the lines of each output field.

10. Apparatus according to claim 9, including means for selecting the weighted contributions to filter the signal with a filter function which has a lower cut-off frequency in the vertical direction for temporal frequencies of half the field frequency and add multiples thereof than for temporal frequencies in the region of zero and even multiples.

11. Apparatus according to claim 10, in which the filter function comprises for each output line taking contributions fran a plurality of lines of the corresponding input field and from a plurality of lines of the adjacent input fields, the contributions from certain lines to either side of the corresponding line in the adjacent field being negative and the contributions from lines between the said certain lines being less than the contributions from the associated lines of the current field.

12. Apparatus according to claim 11 in which the filter function is substantially as shown in Figure 7 or Figure 9 of the drawings.

13. Apparatus according to any of claims 9 to 12 for use when the input video signal is an interlaced signal at a higher number of lines per picture than the output video signal, the apparatus further comprising a pre-filtering means for converting, as a first step, the input video signal to a non-interlaced signal having the said higher number of lines per field.

14. Apparatus according to any of claims 9 to 13 in which the input video signal is at the 1250/50/2 standard and the output videuo signal is at the 625/50/2 standard.

15. Apparatus for generating an output video signal of interlaced form from an input video signal of the same field rate, the apparatus being substantially as herein before described with reference to the drawings.