Digital image processing for converting images from one sampling structure to another
View Patent ↗First image data at a lower sampling frequency is up-sampled in a sampling ratio N:M to a higher sampling frequency in an up-sampling filter; and, second image data at the said higher sampling frequency is down-sampled in a sampling ratio M:N to the said lower sampling frequency in a down-sampling filter where the combination of the up-sampling filter and the down-sampling filter is substantially transparent and every filtered sample is formed from a weighted sum of at least two input samples.
1. Digital image processing apparatus in which first image data representing a digital image at a lower sampling frequency is up-sampled in a sampling ratio N:M to a higher sampling frequency in an up-sampling filter to produce second image data representing the said digital image at the said higher sampling frequency; and, said second image data at the said higher sampling frequency is down-sampled in a sampling ratio M:N to the said lower sampling frequency in a down-sampling filter to produce third image data representing said digital image at said lower sampling frequency, where the combination of the up-sampling filter and the down-sampling filter is substantially transparent characterised in that every filtered sample is formed from a weighted sum of at least two input samples, where N and M are integers, where 1<N<M and where M is not a multiple of N.
2. Apparatus according to claim 1 , in which the combination of the up-sampling filter and the down-sampling filter is substantially transparent in the sense that the reconstruction error due to the combination of the up-sampling filter and the down-sampling filter is smaller than one least-significant bit of the said digital image processing.
3. A method of digital image processing in one or more digital image processors for achieving a transparent cascade on up conversion in the sampling rate ratio N:M where N and M are integers and subsequent M:N down conversion, where the up and down conversion ratios N:M and M:N respectively are rational numbers and the integers N and M satisfy the condition 1<N<M, wherein the up conversion filter operates on a sampled signal S input and is chosen to take the form S up (n)=ΣS input (k).g(Nn-Mk) where k is the running integer over which the sum is taken and wherein a corresponding down conversion filter operates on the up converted signal S up and is chosen to take the form S down (n)=ΣS up (k).h(Mn-Nk); the pair (g, h) of up and down conversion filters being chosen so that Σh(Mn-Nk).g(Nk-Mm) is equal to unity if n=m and is otherwise equal to zero, and wherein every filtered sample is formed from a weighted sum of at least two input samples.
4. A method according to claim 3 in which at least part the said first and second image data represent the same portrayed object.
5. A method according to claim 3 in which the amplitude of the response of the said up-conversion filter and the amplitude of the response of the said down-conversion filter have substantially equal magnitudes at a frequency of half the said lower sampling frequency.
6. A method according to claim 3 in which the frequency response of the said up-conversion filter and the frequency response said down-conversion filter are substantially identical.
7. A method according to claim 3 in which an up- or down-sampling filter has no phase coincidence between its input and output samples.
8. A method of digital image processing in one or more digital image processors for achieving a transparent cascade on up conversion in the sampling rate ratio N:M where N and M are integers and subsequent M:N down conversion, where the up and down conversion ratios N:M and M:N respectively are rational numbers and the integers N and M satisfy the condition 1<N<M, wherein the up conversion filter operates on a sampled signal S input and is chosen to take the form S up (n)=ΣS input (k).g(Nn-Mk) where k is the running integer over which the sum is taken and wherein a corresponding down conversion filter operates on the up converted signal S up and is chosen to take the form S down (n)=ΣS up (k).h(Mn-Nk); the pair (g, h) of up and down conversion filters being chosen so that Σh(Mn-Nk).g(Nk-Mm) is equal to unity if n=m and is otherwise equal to zero, and wherein the frequency response of the up conversion filter is substantially the same as the frequency response of the down conversion filter.
9. A method of digital image processing in one or more digital image processors for achieving a transparent cascade on up conversion in the sampling rate ratio N:M where N and M are integers and subsequent M:N down conversion, where the up and down conversion ratios N:M and M:N respectively are rational numbers and the integers N and M satisfy the condition 1<N<M, wherein the up conversion filter operates on a sampled signal S input and is chosen to take the form S up (n)=ΣS input (k).g(Nn-Mk) where k is the running integer over which the sum is taken and wherein a corresponding down conversion filter operates on the up converted signal S up and is chosen to take the form S down (n)=ΣS up (k).h(Mn-Nk); the pair (g, h) of up and down conversion filters being chosen so that Σh(Mn-Nk).g(Nk-Mm) is equal to unity if n=m and is otherwise equal to zero, and wherein the frequency response of the down converter at the sampling frequency of the sampled signal S input is less than 10%.
10. A method according to claim 9 , wherein the frequency response of the down converter at the sampling frequency of the sampled signal S input is less than 5% of the frequency response at DC.
11. A method according to claim 9 , wherein the frequency response of the down converter at the sampling frequency of the sampled signal S input is less than 1% of the frequency response at DC.
12. A digital image processing system in which first image data at a lower sampling frequency is up-sampled in a sampling ratio N:M to a higher sampling frequency in an up-sampling filter; and, second image data at the said higher sampling frequency is down-sampled in a sampling ratio M:N to the said lower sampling frequency in a down-sampling filter, where N and M are integers, where 1<N<M and where M is not a multiple of N, wherein a frequency response is optimised to minimise the reconstruction error due to the combination of the up-sampling filter and the down-sampling filter and wherein the filter aperture of the up-sampling filter and the filter aperture of the down-sampling filter are each constructed from said optimised frequency response.
13. A system according to claim 12 , wherein in said optimised frequency response the response at the lower sampling frequency is less than 10%, preferably less than 5 % and more preferably less than 1 % of the frequency response at DC.
14. A system according to claim 12 , wherein the reconstruction error is minimised in the sense of being smaller than one least-significant bit of the said digital image processing.
15. A method of digital image processing in one or more digital image processors in which first image data at a lower sampling frequency is up-sampled in a sampling ratio N:M to a higher sampling frequency in an up-sampling filter; and, second image data at the said higher sampling frequency is down-sampled in a sampling ratio M:N to the said lower sampling frequency in a down-sampling filter where the combination of the up-sampling filter and the down-sampling filter is substantially transparent characterised, where N and M are integers, where 1<N<M and where M is not a multiple of N and wherein the frequency response of the down converter at the lower sampling frequency is less than 10%.
16. A method according to claim 15 , wherein the frequency response of the down converter at the lower sampling frequency is less than 5% of the frequency response at DC.
17. A method according to claim 15 , wherein the frequency response of the down converter at the lower sampling frequency is less than 1% of the frequency response at DC.
18. A non-transitory computer program product contained instructions adapted to cause programmable apparatus to implement a method of digital image processing for achieving a transparent cascade on up conversion in the sampling rate ratio N:M where N and M are integers and subsequent M:N down conversion, where the up and down conversion ratios N:M and M:N respectively are rational numbers and the integers N and M satisfy the condition 1<N<M, wherein the up conversion filter operates on a sampled signal S input and is chosen to take the form S up (n)=ΣS input (k).g(Nn-Mk) where k is the running integer over which the sum is taken and wherein a corresponding down conversion filter operates on the up converted signal S up and is chosen to take the form S down (n)=ΣS up (k).h(Mn-Nk); the pair (g, h) of up and down conversion filters being chosen so that Σh(Mn-Nk).g(Nk-Mm) is equal to unity if n=m and is otherwise equal to zero, and wherein every filtered sample is formed from a weighted sum of at least two input samples.