GB2348065A

GB2348065A - Signal compression

Info

Publication number: GB2348065A
Application number: GB9906444A
Authority: GB
Inventors: Derek William Brown; Kevin Alistair Murray; Colin Davies
Original assignee: NDS Ltd; Tandberg Television AS
Current assignee: Synamedia Ltd; Ericsson Television AS
Priority date: 1999-03-19
Filing date: 1999-03-19
Publication date: 2000-09-20
Also published as: GB9906444D0

Abstract

The matching pursuits algorithm is used for the analysis and compression of digital signals, and is used to particular advantage in the compression of motion residual signals in a video encoder. The number of function coefficients is reduced by representing the relatively few large magnitude coefficients by two or more lower magnitude coefficients.

Description

METHOD AND APPARATUS FOR ENCODING A SIGNAL The present invention relates to the field of signal compression, and advantageously the compression of signals using iterative compression techniques, such as matching pursuits.

The matching pursuits algorithm is a computationally complex algorithm used for the analysis and compression of digital signals, and is used to particular advantage in the compression of motion residual signals (error signals) in a video encoder.

Compressing a signal using matching pursuits involves matching a set of predefined'basis signals'to the signal and recording an arrangement of these basis signals, combined with offsets and a magnitude coefficient. This allows the original signal to be substantially reconstructed by a suitable decoder.

Collectively, a basis signal with its associated offset parameters and magnitude coefficient is known as an atom. A compressed signal comprises a sequence of atoms.

The matching pursuits technique is an iterative process. Each time a basis signal is matched to part of the signal to be compressed, the basis function is subtracted from the original signal. The residual signal (the original signal minus the selected basis signal) is then searched again and another basis signal is selected which matches another part of the signal. This basis signal is subtracted from the residual signal, and the process continues until the original signal has been substantially encoded.

To maintain the quality of the encoded signal, the correlation results need to be accurately represented. This is usually achieved by allocating a large number of bits for each correlation result. The number of bits required to accurately represent the correlation results is usually referred to as the bitprecision. The size, cost, complexity and performance of apparatus to implement matching pursuits is directly related to the bit-precision required. It is therefore desirable to minimise the size, cost and complexity of such apparatus, whilst maintaining high performance.

Quantisation of the correlation results can be used to reduce the bit-precision, however quantisation reduces the range of available correlation results which may in turn significantly reduce the quality of the encoded signal.

Accordingly, one aspect of the present invention is to reduce the complexity of a matching pursuits encoder, without unduly affecting the quality of an encoded signal.

According to a first aspect of the present invention, there is provided a method of encoding a signal using matching pursuits in which the signal, having a predetermined magnitude range, can be represented by a function and an associated function coefficient, the method comprising the steps of: selecting an optimum function and associated function coefficient from a predetermined range of function coefficients to optimally encode a portion of the signal ; characterised in that the selecting step further comprises limiting the range of function coefficients ; and selecting the optimum function coefficient from the limited range of function coefficients to encode a portion of the signal within the magnitude range.

According to a second aspect of the present invention, there is provided apparatus for encoding a signal using matching pursuits in which the signal, having a predetermined magnitude range, can be represented by a function and an associated function coefficient, the apparatus comprising: a selector for selecting an optimum function and associated function coefficient from a predetermined range of function coefficients to optimally encode a portion of the signal ; characterised in that the selector limits the range of function coefficients; and wherein the selector selects the optimum function coefficient from the limited range of function coefficients to encode a portion of the signal within the magnitude range. One advantage of this, is that the size of the hardware required to implement the present invention is reduced.

The invention will now be described, with reference to the accompanying drawings, in which: Figure 1 is a diagram showing a typical example of how matching pursuits is used to compress a signal ; Figure 2 is a diagram showing an embodiment of a typical matching pursuits compression engine ; Figure 3 shows a series of graphs representing a typical statistical distribution of atom coefficients.

Figure 4 is a diagram showing an example of how the present invention implements matching pursuits to compress a signal ; Figure 5 is a graph showing the statistical distribution of atom coefficients after processing according to the present invention; Figure 6 is a diagram showing a first embodiment of the present invention; Figure 7 is a diagram showing a second embodiment of the present invention; and Figure 8 is a diagram showing a third embodiment of the present invention.

Figure 1a shows a signal 100 which is to be compressed using matching pursuits. An atom 110 is determined as being the best match for a portion 105 of the signal 100. The atom 110 is subtracted from the signal 100 to give a residual signal 120 shown in Figure 1b. It can be seen in Figure 1b that the peak present in the portion 105 of the signal 100 has been substantially removed, to leave the residual signal 120.

Figure 2 is a diagram showing an embodiment of a typical matching pursuits compression engine. A signal to be compressed 201 is input to a data store 210. The stored signal is then input to a correlator which performs the matching pursuit technique on the stored signal. The correlator matches basis functions to the stored signal, and stores a series of results for each of the matches. An indication as to the accuracy of the match is recorded by multiplying each point on the basis function by each point on a section of the stored signal. The higher the resulting coefficient, the more accurate the match. Due to the number of matches performed by the correlator, millions of accuracy coefficients are stored. The results are then analyse by a controller 212 which finds the highest accuracy result from the stored coefficients. The chosen atom is then subtracted from the data store 210 and the process continues. The chosen atom is stored in an atom store 213. The data in the atom store 213 may be subsequently statistically encoded by a Huffman or other statistical encoder 214. A resulting compressed output signal representing the input signal 201 is output at a terminal 215.

Figure 3 shows a series of graphs representing a typical statistical distribution of atom coefficients after processing according to the conventional matching pursuits technique described above. It can be seen in Figure 3a that relatively few large magnitude atom coefficients occur per frame, with the majority of atoms selected having low magnitudes. The present invention recognises and exploits this statistical distribution.

For an input signal having a given magnitude range each atom coefficient is allocated a sufficient number of bits (bit-precision) such that the entire range of input signals can be represented by a single atom coefficient. For the purposes of illustration, assume the typical bit-precision is 10 bits, which means that atom coefficients from either 0 to 1024, or from-512 to +512, can be represented. Given that the majority of atom coefficients are of relatively low magnitude, it can be seen that the allocation of 10 bit bit-precision is somewhat inefficient, since the majority of atom coefficients could be represented in a much lower number of bits.

Due to hardware implementation constraints, it is not desirable to employ a variable length coding scheme.

The present invention is based upon the realisation that the range of atom coefficients can be reduced by representing the relatively few large magnitude coefficients by a combination of two or more lower magnitude coefficients. By selecting a suitable upper limit, all atom coefficients above that limit are decomposed into combinations of lower value coefficients as illustrated in Figure 3b.

The result of this potentially increases the number of atoms required to represent a given signal but reduces the number of different coefficients used as all high value coefficients are represented as a combination of lower value coefficients. A statistical encoder, such as an arithmetic or Huffman encoder, exploits the statistical distribution of the atoms coefficients and is able to achieve greater compression ratios than would be possible if an upper limit was not used.

Referring now to Figure 4, there is shown an example of how the present invention implements matching pursuits to compress a signal. A signal to be compressed 100 is shown. An ideal atom 110 is determined as being the best match for a portion 105 of the signal 100, however the magnitude of the ideal atom 110 exceeds that of predetermined clipping threshold 301. The ideal atom 110 is therefore clipped to produce a clipped atom 310 which is below the clipping threshold 301. When the clipped atom 310 is subtracted from the input signal 100, the residual signal 319 is left. It can be seen that the portion 105 of the signal 319 (Figure 3b) has not been completely encoded, since there is still a signal component remaining. The matching pursuits technique is performed again on this section. This time an ideal atom 312 is chosen. The ideal atom 312 is below the clipping threshold 301 so no clipping of the atom 312 is required. This atom is subsequently subtracted from the signal 319 which results in residual signal 320 (Figure 3c). It can be seen that the portion 105 of the residual signal 320 has now been substantially encoded using two atoms.

The method of the present invention can lead to greater compression, since a narrower range of atom coefficients can be better exploited by a statistical encoder. However, greater compression ratios is only one advantage of the present invention.

A further advantage of the present invention is in the corresponding reduction in the bit-precision required. Since the range of atom coefficients is reduced by the use of an upper limit, so a reduced number of bits is required to represent all possible atom coefficients.

If 10 bit precision is typically used, this allows atom coefficients from 0 to 1023 to be represented. If an upper limit for atom coefficients of, for example, 511 is imposed, this means that all atom coefficients can now be represented in 9 bit bit-precision (29 = 512), giving a 10% saving in the number of bits required.

Figure 5 is a graph comparing a typical statistical distribution of atom coefficients with and without the use of clipping. A curve 501 shows the distribution of atom coefficients for a typical conventional matching pursuits encoder without the use of clipping. A curve 500 shows the distribution of atom coefficients for a typical matching pursuits encoder using the method of the present invention. It can be seen that, in the conventional matching pursuits encoder a small number of high magnitude atom coefficients are present above a predetermined upper limit 502. In the matching pursuits encoder of the present invention, all the atom coefficients are below the upper limit 502.

Figure 6 shows a block diagram of one embodiment of the present invention.

The system shown in Figure 6 is functionally similar to the system shown in Figure 2 but with the addition of a clipping circuit 220.

Compared to the system of Figure 2, the position of the clipping circuit in the system of Figure 6 allows the size of the atom store 221 to be reduced (compared the system of Figure 2 without a clipping circuit) since a reduced number of bits is required to represent the range of possible atom coefficients.

The system of Figure 7 is a further adaptation on the system of Figure 2. A clipping circuit 230 is positioned before the atom coefficient selector 231. In this embodiment, due to the position of the clipping circuit 230 the amount of hardware required to implement the atom coefficient selector 231 is reduced as well as the size of the atom store 232.

The system of Figure 8, is a further adaptation on the system of Figure 6 and Figure 7. In this embodiment, the clipping circuit is incorporated into the correlator 240. This embodiment can lead to further reductions in the amount of hardware required, since the size of the correlator can be reduced, as well as the size of the atom coefficient selector 241 and the atom store 242.

By careful selection of the upper limit for use in the clipping circuit the required bit-precision can be reduced without affecting image quality. The savings in bit-precision relate directly into savings in the hardware implementation of such a system.

In a further embodiment of the present invention the upper limit is varied in operation, for example between different types of encoded video frame. In yet a further embodiment the upper limit is continuously adapted, for example in dependence on the image content.

Although the present invention has been described with reference to matching pursuits techniques, it will be appreciate by those skilled in the art that the method of the present invention can equally be applied to other iterative encoding techniques.

Claims

CLAIMS 1. A method of encoding a signal using matching pursuits in which the signal, having a predetermined magnitude range, can be represented by a function and an associated function coefficient, the method comprising the steps of: selecting an optimum function and associated function coefficient from a predetermined range of function coefficients to optimally encode a portion of the signal ; characterised in that the selecting step further comprises limiting the range of function coefficients; and selecting the optimum function coefficient from the limited range of function coefficients to encode a portion of the signal within the magnitude range.
2. The method of claim 1, further comprising selecting a plurality functions and associated coefficients to optimally encode a portion of the signal where a portion of the signal cannot be encoded using a single function and associated function coefficient.
3. The method of claim 1 or 2, further comprising setting the limit for the limited range of function coefficients to a predetermined level.
4. The method of claim 1,2, or 3, including supplying the signal as a video signal comprising encoded frames.
5. The method of claim 4, further comprising varying the limit for the limited range of function coefficients according to the type of encoded video frame.
6. The method of claim 4, or 5, further comprising varying the limit for the limited range of function coefficients according to the image content of an encoded frame.
7. The method of any preceding claim, further comprising statistically encoding the encoded signal.
8. Apparatus for encoding a signal using matching pursuits in which the signal, having a predetermined magnitude range, can be represented by a function and an associated function coefficient, the apparatus comprising: a selector for selecting an optimum function and associated function coefficient from a predetermined range of function coefficients to optimally encode a portion of the signal ; characterised in that the selector limits the range of function coefficients; and wherein the selector selects the optimum function coefficient from the limited range of function coefficients to encode a portion of the signal within the magnitude range.
9. The apparatus of claim 8, wherein the selector is adapted to select a plurality functions and associated coefficients to optimally encode a portion of the signal where a portion of the signal cannot be encoded using a single function and associated function coefficient.
10. The apparatus of claim 8 or 9, wherein the selector sets the limit for the range of function coefficients to a predetermined level.
11. The apparatus of claim 8,9, or 10, wherein the signal is a video signal comprising encoded frames.
12. The apparatus of claim 11, wherein the selector sets the limit for the range of function coefficients according to the type each encoded video frame.
13. The apparatus of claim 11, or 12, wherein the selector sets the limit for the limited range of function coefficients according to the image content of each encoded frame.
14. The apparatus of any preceding claim, further comprising statistically encoding the encoded signal.