GB2117209A - Video processing systems - Google Patents

Abstract

An input video frame comprises low definition pixels (K, L, M, N frame 1 and Q, R, S,T frame 2) at a low frame rate. The input frame is stored in a high definition frame store which is recursively updated and filled using spatial interpolation (Fig. 8). e.g. to give a value at W, and temporal interpolation (Fig. 9) to give values between the input frames and enable an output at a higher definition and higher frame rate. <IMAGE>

Description

SPECIFICATION Video processing systems The invention relates to a video processing system inter alia for use in special effects in television.

In known systems used for special effects for example video information ; received by a framestore acting as a buffer and in order to produce the special effect desired, an output processor typically receives data from selected addresses within the framestore to reconstitute a picture of different shape or size to that input to the store.

Whilst such systems work over a limited range of special effects as currently available, the way in which they operate inhibits their flexibility.

An object of the invention is to provide a system capable of producing greater flexibility in picture manipulation whilst maintaining picture quality so that the resultant picture is not noticably degraded.

According to the invention there is provided a video processing system for picture shape manipulation comprising:- frame storage means for receiving picture point information in a plurality of locations equivalent to a video frame; addressing means for addressing selected frame store locations a plurality of times within a frame period in dependence on the manipulation required; processing means for processing the picture information at any given location each time that location is addressed; and control means for varying the processing provided in dependence on the density of the picture information at a given picture location within a frame period.

According to a further aspect of the invention there is provided an addressing mechanism for a framestore processing system comprising: means for determining only selected ones of the desired framestore addresses required to be accessed; means for updating the selected ones of the desired addresses at a rate slower than normal frame rate; and address interpolation means for calculating all the desired addresses to be accessed from the available address data in both spatial and temporal modes so that all the addresses to be accessed are available updated at normal frame rates.

The invention will now be described with reference to the accompanying drawings in which: Figure 1 shows a basic embodiment of the system relating to one aspect of the present invention; Figure 2 shows various aspects of picture manipulation; Figure 3 shows the processing aspects of the system of Figure 1 in more details; Figure 4 shows further details of the processing; Figure 5 shows a further aspect of picture manipulation; Figure 6 shows a system capable of providing suitable address manipulation; - Figure 7 shows address interpolation techniques for the system relating to a second aspect of the present invention; Figure 8 shows an address interpolation configuration for dealing with the spatial manipulation; and Figure 9 shows an arrangement for both spatial and temporal interpolation of the addresses.

A basic embodiment relating to one aspect of the invention is shown in Figure 1. A processor 10 receives incoming picture information together with information previously stored in framestore 11. The x, y address accessed from the framestore designating a given picture piont is determined by the output of address mechanism 12. In addition a control parameter Z is provided by the address mechanism for controlling the processing within processor 10.

The arrangement shown is a radial departure from known systems in that the address mechanism in effect defines the 'shape' of the image to be produced by a process which builds up a picture by accessing certain framestore locations more than once within the frame period so that effectively all the original picture points are arranged to go somewhere even though the locations will be different to their original pixel positions. In addition the address mechanism produces the control parameter to ensure that correct fractions of the picutre points are added to the framestore.

The processing and addressing mechanism will now be explained in more detail. The address and storage mechanism is more complicated than at first apparent in that we have devised a system whereby the picture points can be considered as part of a grid of cells and that these picture points need not be assigned solely to one cell but can have portions assigned to a number of cells as represented in Figure 2. Thus a standard picture (without processing) made up of a plurality of picture points would be stored in successive cells within the framestore as represented by picture points P1, P2, P3 and P4of Figure 2(a), and the address mechanism 12 can be considered as producing a standard addressing sequence with the value of Z being fixed typically at a value equal to 1.Although only pixels P1 to P4 are shown it is clear that all the cells of the framestore would contain a pixel in this mode.

The representation of Figure 2(b) shows the change in the picture points, where a picture (still of standard size) is scrolled horizonally and vertically by + a picture point. For simplicity only P, and P2 are shown. The pixels P, and P2 can now be seen to each have four portions allocated to adjacent store cells. In practise the portion of a given pixel added to a cell is known from the x and y address provided by our address mechanism 12, which is arranged to produce a main address and remainder, the remainder giving the fraction allocated to a given cell and comprise the Z parameter.

Thus for Figure 2b, the framestore 11 and processor 10 under the control of address mechanism 12 effectively initially receive pixel P, and after processing partially assign it to cells C and C2 and the appropriate cells on the next line Cq and Cr. Pixel P2 is then processed and placed in C2, C3, Cr and Cs. The process in practice takes into account the information already alloted from pixel P, in cell C2 for example so that this cell location is accessed for a read operation followed by the processing dependent on the value of Z provided and then the processed output is written into the store.

In practice where scrolling takes place without picture size change (zoom) then the portions of the adjacent pixels allocated to a cell will always add up to 1 at the end of the processing. In the Figure 2(b) illustration, four read, processing and write operations will be required to produce the desired picture information with Z in this example being $.

An embodiment of the processor 10 suitable for providing the necessary basic picture manipulation is shown in Figure 3.

A multiplier 13 receives the incoming information and after multiplying the information, this is passed to adder 14. The value of Z will always be chosen to be between 0 and 1.

In the example of 2(a) above as already explained, the value of Z can be considered as equal to 1 and in the 2(b) arrangement it will be T since four quarters from adjacent pixels are used to generate the information within cell C2. Thus providing the contents of the store is cleared to zero at the start of each picture the interpolation for generating the information for each cell is automatically achieved merely by the succession of read-process-write operations.

However, this mechanism is only suitable for scrolling as it stands, when there is 2 change in picture size then additional manipulation is required to avoid errors in picture build up, as now explained with reference to Figure 2c. Here a 2:1 reduction in picture size is represented (without scroll) so that pixel P, is within cell C, and P2 is partially within cell C, and C2. Similarly P3 is wholly within cell C2 and P4 is partly within C2 and C3. If some adjustment of the data quantity were not made then the resultant information within cell C2, for example, would be twice that desired (considering picture intensity, for example). Thus it is necessary to provide the means for adjusting this situation and from Figure 2(c) it can be seen that the adjustment value (K) is required to be T to provide the desired adjustment at each cell.

Thus although this is a simple case, the general rule in fact follows this explanation in that K=compression ratio. Compression ratios of 32:1 have been successfully achieved. Although K can be considered as fixed for a given compression (e.g., K=3 for 3 times compression) the compression need not be the same across the entire picture and indeed variable compression gives rise to a host of special effects which can be achieved with the present system.

Although the mechanism for (read-add manipulated new picture point-write) in a single cycle may initially seem straightforward, in fact it is a very powerful tool which performs interpolation and filtering all in the one operation or sequence of operations without the need of elaborate additional devices.

Considering Figure 2(d) it can be seen that compression and scrolling can be achieved by moving the addressing by T pixel (only P, and P2 are shown for simplicity).

Although Figure 3 shows the basic mechanism involved, it has been somewhat simplified and more comprehensive system for producing the desired processing is shown in Figure 4 and shows how the fractional part of the address and the density compensation referred to above is used. A multiplier 20 is now included to provide the density compensation. As illustrated in Figure 2(b) a pixel may be manipulated so that four portions are allotted to each of four adjacent store cells. In order to cope with this manipulation at reasonable speeds it is typically necessary to include additional processing and storage over the mechanism as in Figure 3 where the 4 points would have to be computed sequentially during one input pixel.The single multiplier, adder and framestore of Figure 3 has been replaced by four framestores 34 to 37 each with their associated multipliers 30-33 and adders 1 6 to 1 9 respectively. The outputs from the framestores are received by summer 38 to produce the combined output. Such a system allows the incoming pixel to be available to each of the four relevant store cells and their associated processing. The address mechanism 12 is now shown as producing the main x, y address for the four respective framestores (A, B, C and D) and in addition the fractional (Z) part of the address (a, b, c, and d) and the density compensation value (K).

The main address for pixel P, for any of the illustrated situations will be A=C1, B=C2, C=Cq and D=Cr. The fractions will vary. Thus for Figure 2(a) a=1 and bc and d=0, and K=1.

For Figure 2(b) when dealing with P1, the fractional addresses a, b, c and d=, and K=1.

Any combination of these portions will always equal 1 due to the presence of summer 38.

For the Figure 2(c) situation, for pixel P,; a=1 and b, c and d=O and K=2F, for pixel P2; A, B, C and D will be as before but a and b=+ and c and d=O with K=+.

It is to be remembered that each store 34 to 37 is a complete framestore. Thus when writing into the system the addresses always express four different but adjacent cells, but on readout from the system, the address for each of the framestores is the same. In other words, for processing you want to access 4 cells, but on readout you only want to access a single cell.

The provision of multiplier 20, effectively reduces the incoming data, which is required where compression is taking place otherwise a build up of contributions from many picture points into a cell would give too large a density of information. For special effects the compression will not be the same for each cell within the frame.

For clarity multiplier 20 has been shown separate from multipliers 30-33 but in practice multiplier 20 could typically be incorporated within divider 30, 31, 32 and 33.

Another special effect will now be described to illustrate the versatility of the system. Figure 5 shows an effect equivalent to page P being turned. Merely generating the 'shape' you want allows the correct sequence of addressing to be achieved as well as the correction for the build up of the picture. Thus at the edge E of page P there will be a greater build up (but compensated by multiplier 20, than in the overlap portion F).

In practise the flap F will appear to be transparent so that picture information underneath will also be visible.

If it is desired to make the flap opaque so that the underlying picture is obscured this can be achieved by the operational cycle of read-replacewrite, which can be simply produced by inhibiting the connection between the framestore 11 output and the adder 14 in the Figure 3 configuration.

Typically it is convenient to generate an identification 'tag' via the address mechanism to ensure that the system knows which area is above the other area.

The way in which the address mechanism 12 can operate for any shape desired is now shown in more detail in Figure 6.

It can be seen from the above examples that the address sequence chosen by the operator effectively defines the shape and size of the output picture as well as compensating for the accumulated information by the provision of x, y, z and K parameters.

The simple shapes of Figure 2 can easily be generated by the keyboard 21 for input to computer 20 so as to provide the sequence of cell address locations desired to be accessed as well as entering the value of Z and K required.

Thus the grid of framestore cells can be determined to be within the boundary of the desired picture shape or not, when they are, these addresses are accessed during manipulation.

Alternatively, standard mathematical formulae can be entered to generate the desired shapes in the computer 20. In the case of a circle for example, the standard text book equation for the circle is entered and simply by defining the cell address of its centre and the circle radius, then it is possible to determine whether a particular cell address is within the boundaries of the circle or not, and this defines the resultant picture shape.

The desired shape area is also an indication of the compression relationship and so this also can be calculated to determine the K value. The value of compression (K) for a given picture area can be determined for example by employing standard area computational techniques (see p. 129-131 of Hewlett Packard HP25 handbook 1975) where K is proportioned to the area.

The calculated values for x, y together with the appropraite value for K for that cell for a given shape are then passed to disc store 23. In practice disc 23 contains a whole range of shapes including shape sequences to allow picture transformation from one shape to another to be achieved. Once the shapes have been entered the computer 20 is not necessarily required thereafter, and the system merely uses the disc store 23 as its shape data source. Although the disc access is much faster than that from the computer, it nevertheless is generally not rapid enough to cope with video rates of addressing. In order to overcome this problem we have incorporated an additional mechanism represented by address interpolator 24 which operates as illustrated in Figure 7.The disc store in practice only holds coarse cell address data as shown by points K, L, M, N of which K and L represent typically the 1 st and 8th successive pixel address horizontally and M and N the equivalent address points 8 lines below. Points Q, R, S and Tare typically the equivalent address points 8 frames later. Thus updating the addressing at this rate can be handled by the disc and the addresses between the available points are interpolated therefrom both spatially and temporally as illustrated. We have found that this technique does not produce any noticeable degradation to the picture produced.

Although the computer 20 has been described as providing all the values of x, y, Z and K, where only coarse addresses are provided to disc 23, then it may be convenient to only provide correspondingly coarse values for the other parameters and then these also are interpolated to derive all the desired information. Alternatively, the parameters can be calculated following the address interpolation process within interpolator 24, using the computation referred to above.

The address interpolation technique also works where the disc is producing an effects sequence and whilst the change in addressing is produced by the disc updated every 8 frames (in this example), the address interpolation produces a graduai change over the 8 frames, by giving greater weighting to the adjacent frame than the remote frame.

An arrangement for providing the spatial address interpolation is shown in Figure 8. The coarse addresses are received by address delay latch 41 which provides a delay equivalent to 8 lines of address. The delayed address is passed to multiplexer 42 and the current coarse address is passed to multiplier 43 before addition in adder 44. The adder output passes to a further delay 45, which has a delay equivalent to 8 picture point addresses and this delayed output passes to multiplier 46. The undelayed output from adder 44 passes to multiplier 47 prior to receipt by adder 48, which also receives the output from multiplier 46.

The Figure 8 arrangement in practice is duplicated to give the necessary interpolation for both the x and y address.

Thus from Figure 7, as the K, L, M, N coarse addresses are available, any other interpolated address e.g., address W can be determined therefrom. The value of k and I will vary between 0 and 1 typically in + steps as the addresses are calculated. These values for the multipliers can conveniently be provided by look up tables incremented by the address clocks.

The spatial address interpolator is incorporated in the Figure 9 arrangement to produce the temporal interpolation. The output from the disc store 23 is for ease of explanation shown as being provided from a first shape store 23A (holding the KLMN addresses of Figure 7 for example) and a second shape store 23B (hoiding the QRST values). After spatial interpolation, the address values are then available for temporal interpolation usings multipliers 52 and 53. The resultant outputs are available via adder 54. The values fort again vary between 0--1 in + steps.

conveniently using a look up table 55. This allows any change in address shape between the 8 frame period to be gradually introduced.

Although the system has been described as having coarse addresses operating around 8 addresses and 8 frames, this value is not mandatory.

In practice although the system has been generally described in relation to handling intensity values for the video information, when handling colour data the system would typically be triplicated so that one part handles the luminance data and the other part the chrominance (e.g., colour difference) information.

In the N.T.S.C. system this is coded as Y, I and 0 information respectively. From this it can be seen that although a relatively large number of framestores are needed, the results obtained justify such a configuration.

Alternatively the colur can be handled on a RGB basis.

Although the system has been described for use in special effects for broadcast T.V. it can be used for other types of video systems requiring picture manipulation in a total free form after generation.

Claims (12)

Claims

1. An addressing mechanism for a frame store processing system comprising:- means for determining only selected ones of the desired frame store addresses required to be accessed, means for updating the selected ones of the desired addresses at a rate slower than normal frame rate, and address interpolation means for calculating all the desired addresses to be accessed from the available address data in both spatial and temporal modes so that all the addresses to be accessed are available updated at normal frame rates.

2. A mechanism as claimed in Claim 1, wherein said determining means includes a memory for retaining at least one addressing sequence.

3. A mechanism as claimed in Claim 2, wherein said updating means is adapted to access different portions of said memory.

4. A mechanism as claimed in Claim 3, wherein said memory and said updating means is provided by a movable recording medium containing only said selected frame store addresses from selected frames.

5. A mechanism as claimed in Claims 2, 3 or 4, wherein generator means are provided to determine desired addressing sequences prior to receipt by said memory.

6. A mechanism as claimed in any one of Claims 1 to 5, wherein said address interpolation means includes a first spatial address interpolator for determining all the relevant addresses from the information available associated with a first frame, a second spatial address interpolator for delivering all the relevant addresses from the information available associated with the next available frame, and a temporal address interpolator for delivering the relevant addresses associated with intermediate frames from information provided by both said first and second spatial address interpolators.

7. A mechanism as claimed in Claim 6, wherein each of said spatial address interpolators includes an arithmatic processor for variably providing synthesised addresses in dependence on their spatial relationship with said address information currently available.

8. A mechanism as claimed in Claim 7, wherein said temporal address interpolator includes an arithmetic processor for variably providing synthesised address information in dependence on their temporal relationship with the address information currently available, whereby the addresses within one frame are modified over intermediate frame periods to effect a gradual change from frame to frame.

9. A mechanism as claimed in Claim 7 or 8, wherein an arithmetic information generator is provided to variably control the desired synthesis on a temporal basis.

10. A method of providing addresses for a frame store processing system comprising the steps of:- providing only selected ones of the desired frame store addresses required to be accessed, updating the selected ones of the desired addresses at a rate slower than normal frame rate, calculating all the desired addresses to be accessed by interpolating the available address information both in spatial and temporal modes so that all the addresses to be accessed are available updated at normal frame rates.

11. An addressing mechanism for a frame store processing system substantially as described herein and with reference to the accompanying drawings.

12. A method of providing frame store addresses substantially as described herein.