DE4128977A1 - Video signal filter circuit for transformed and quantised data blocks - uses two=dimensional block coordinates with relative coordinates or transformation coefficients with two=dimensional variables - Google Patents

Abstract

The circuit filters a video signal represented by sampled values of pixels (I(q,x)) whose geometrical location in a video picture are determined by two-dimensional block coordinates (q) and relative coordinates (x). Alternatively, the signal is represented by transformation coefficients (C(q,u)), u being a two-dimensional variable in the transformation domain. A filtered signal is determined for which the value of the function F is at a minimum, where: F = Sum(x,q) (IS(q,x)-I(q,x))power2 + mu.Sum(q,x) (IS(q,x) - IP(q,x))power2 and the difference between the transformation coefficient (CS(q,u)) for the filtered video signal and that (C(q,u)) for the video signal (I(q,x)) is below a threshold D(q,u). Mu is an experimental parameter. The sum in the second term of the function F uses values of x which belong to the set of edge points (Rq) of the block having the block coordinates q. IP(q,x) is a target boundary value for the boundary point x of the block. The target value is determined from an average value of the filtered video signal, derived from the sampled pixel values which lie adjacent to the block having the coordinates q. ADVANTAGE - Block structure is almost completely suppressed and compression is assisted.

Description

The invention relates to a circuit arrangement for filters of a data block transformed and quantized th video signal, which is determined by the samples I (q, x) Pixels is shown, according to the generic term of claim 1.

A comparable circuit arrangement is e.g. B. in the DE 39 17 085 A1 described. Such circuit arrangements are used for the post-processing of video signals used by a so-called hybrid encoder encoded in blocks, then transmitted or stored and then again in blocks from a hybrid decoder were decoded. The exact mode of operation of such a code rer or decoder, which according to a in the CCITT recomm H.621 defined procedures work is z. B. in the patent applications DE 36 13 343 A1, DE 36 20 424 A1 and DE 36 38 128 A1 described.

With such an encoding, the signal of a video picture structured according to blocks. Blocks are e.g. B. all Data of the pixels in a video image that are on a rectangular screen or square cutout. In the present Case is an example of square cutouts, which are composed of 8 * 8 pixels. The entire Video image exists e.g. B. from 36 block lines and 44 block columns.

To identify a block within a videobil block addresses or block coordinates are used,  which is represented here by a two-dimensional vector q = (q1, q2) with positive, integer components the. The first component of the vector q gives the lines address and the second the column address.

To identify the pixels within an 8 * 8 Blocks become relative coordinates or relative addresses (Coordinate and address are understood here as synonyms den), which is also used here as a two-dim sional vector x = (x1, x2) with integer components being represented. x1 and x2 can be independent of each other others assume the values from 1 to 8.

According to the CCITT recommendation mentioned above, the data of such a block of a two-dimensional cosine Subjected to transformations, d. i.e., by a linear comm combination of 64 cosine functions represented, which are distinguish among themselves by their "frequencies". The total image information of a block with the address q is then in the 64 transformation coefficients C (q, u), where the two-dimensional vector u = (u1, u2) with integer components of both the counting of the coefficients and also the cosine functions (basic functions). C (q, u) is also used to represent the video signal I (q, x) designated in the transformation area with the variable u.

Typically, the components of the vector pass through the numbers from 0 to 7. The Com components of the vector u the "frequencies" (more precisely: waves pay because it is the description of the area Structures and not temporal structures) Cosine functions that are used to describe the areas structures within a block are required are.

The advantage of displaying a video signal through the  Transformation coefficient is that with it informa ions can be compressed. There is z. B. a block from identical pixels, is instead of the transfer of 64 pixel data only the transmission of the coefficient C (q, O) with O = (0, 0) required as zero vector. By the representation of the video signal in the transformation area is the correlation of its values in the local area carried.

To further save data when coding the coefficients C (q, u) quanti by a quantizer siert. The associated quantization error leads however, in the decoded video signal I (q, x) which is based on a Screen is played back, so-called "blocking". Blocking means image disturbances that occur in one Video image more or less clear - depending on the fineness the quantization and depending on the image content - that of the Coding can be used to identify the block structure.

To reduce this disruptive block structure, known circuits (see above) smoothed the block edges tet, e.g. B. by linear filtering of all pixels Video image that is near the block edges.

The invention has for its object a circuit order of the type mentioned, with which the block structure mentioned is almost completely underneath lets press. This task is characterized by the following features solved:

• - Means with which it is provided to determine a filtered video signal IS (q, x) by which the value of the functional under the secondary condition that the difference between a transformation coefficient CS (q, u) for the filtered video signal and the transformation coefficient C (q, u) of the video signal I (q, x) is moderately less than or equal to a threshold D ( q, u), where
• 1) µ is an experimentally determined parameter,
• 2) the sum in the second term for the functional F only runs over the relative coordinates x that correspond to the set Rq the edge points of the block with the block coordinates q belong,
• 3) IP (q, x) a target boundary value for the boundary point x of the Blocks with the block coordinates q is
• - Means intended for this, the target margin IP (q, x) from an average of the filtered video i gnals to be determined, averaged over the samples of Pixels leading to the block with the block coordinates q are neighboring.

The functional F in (1) consists of two parts, namely a first term, in which the sum over all Ortskoor dates of one block and a second term, in which the sum only over the coordinates of the edge points of a Blocks running. The first term should therefore be area term of F and the second term - without the parameter µ as a factor - Boundary term of F.

The marginal term is indefinite insofar as it contains a target margin value IP (q, x), which is not explicitly specified and which is only required to be an average of the sought signal IS (q, x), averaged over image points of the Block q adjacent blocks. Here, mean is understood to mean both a linear, weighted mean with possibly adaptive weights and a non-linear mean. In particular, IP (q, x) can only consist of a value of IS (q ₁ , x), with q ₁ meaning a block adjacent to block q. The boundary term of the sought signal IS (q, x) is intended to be matched to one another.

If the target margin value consists of a linear mean of Boundary values of a neighboring block, this is the minimum of F determine exactly if it is dependent on the constraints will see. The filtered signal IS (q, x) then results than the original signal, with the difference that one The marginal value of a block has been replaced by a with tel value of this block from boundary points and the neighboring block ke. Then, in general, the constraints are violated, which ensure that in the filtered signal IS (q, x) no larger errors than those that occur by quantizing the transform coefficients have been caused in the coding process.

Taking the constraints into account are exact Solutions to the underlying problem are no longer without more available. You have to rely on approximations either on iteration methods or on power series windings z. B. according to the parameter µ.

In both cases, it is recommended to choose the target margin Approximations for the filtered signal to determine the reduce mathematical effort.

If you put the filtered signal IS (q, x) as a linear comm combination of the basic functions, which also applies to the Kodie tion process have been used, F must with respect to the  sought coefficients CS (q, u) made as small as possible will. Because the inaccuracy (due to the quantization at the coding) of these coefficients in the video signal I (q, x) is the cause of the blocking, the possibility arises speed, only those coefficients in the filtered signal approximated to determine which is most responsible for blocking tragic to make a big impact with as little effort as possible to obtain.

Coefficients, their contribution to blocking in principle is negligible, can be formulated by the Supplementary condition in the form D (q, u) = 0 when looking for one Solution are disregarded.

Based on the figure and an embodiment, the Invention will now be explained in more detail. The figure shows a Basic circuit diagram of an arrangement according to the invention.

In the figure, a video signal I (q, x), which has been coded and decoded again in accordance with the H. 261 recommendation, is fed on a line 1 a to a circuit arrangement according to the invention for post-processing. The video signal is structured according to 8 * 8 blocks. The vector g = (q1, q2), with its positive, integer components, indicates the addresses of a block within a video image. The uppermost, leftmost block of a video image displayed on a monitor has the addresses ( 1 , 1 ) and the bottom, rightmost block has the addresses ( 36 , 44 ) because the entire image is made up of 36 * 44 blocks is.

Within a block, the vector x = (x1, x2) specifies the addresses or coordinates of one of the 64 pixels. The pixel that is in a block at the top left has the address ( 1 , 1 ) and the pixel that is in the block at the bottom right has the address ( 8 , 8 ).

The input signal 1 a is fed to an image memory 2 and initially stored there. At the same time, it arrives at a unit 1 which changes the data of each pixel with the block coordinates q and the relative coordinates x by a value which corresponds to a linear combination of those basic functions which were also used in the coding to transform the pixel data. In the present case, the transformation is a cosine transformation and therefore the base functions are cosine functions. The coefficients of the linear combination are chosen randomly between two permissible (see below) limit values, ie, the given signal I (q, x) becomes a new signal IS ₀ (q, x) according to the relationship:

calculated.

In (2) sym is at the summands under the sum symbol Bolically recognizing that it is the basic functions are cosine functions as described in the H.261 recom are set. More precise is under the expression CosB (u, x) with a normalization factor K the size

Cos B (u, x) = K cos (πu1 (2 × 1 + 1) / 16) cos (πu2 (2 × 2 + 1) / 16) (3)

to understand, where u1 and u2 are the two positive, entirely number components of the vector u = (u1, u2) are with where the basic functions are counted. The whole Numbers u1 and u2 run independently of each other Values from 0 to 7. The vector x = (x1, x2) already gives them relative coordinates or addresses of the Pixels within a block.

Since I (q, x) can also be represented as a linear combination of the basic functions CosB (u, x), (2) means that each transformation coefficient of the given signal I has been changed to δ (q, u) ₀ if necessary. The amount of this change must not be greater than the amount of the maximum quantization error that was made for the corresponding coefficient during coding.

The signal IS ₀ (q, x) is stored in a second image memory 3 , specifically as a zero approximation for an iterative process for determining the smallest possible value of the functional F according to formula (1). It is extremely important to note the secondary conditions mentioned, namely to make the magnitude of the difference in the transformation coefficients between the sought signal IS (q, x) and the given signal I (q, x) no greater than a threshold D ( q, u). This requirement also applies to any approximation of the filtered signal. The positive thresholds D (q, u) must also not be greater than the amount of the maximum quantization error that was made during the quantization of the transformation coefficients during the coding.

The minimum of the functional F under the mentioned side Conditions are approximated in the example after the now explained iteration method.

The filtered or searched video signal IS (q, x) also leaves present themselves as a linear combination of the basic functions; it is

with CS (q, u) as the transformation coefficient of the filtered th video signal for the block with the address q. These Coefficients are now viewed as searched quantities; the first step is to act as if it were the minimum  of the functional F without the mentioned constraints determine. It follows that the partial derivative of F must be at least zero according to the searched coefficients got to. The zeros of this derivative are then after the Newtonian approximation. Indeed is initially only the Koeffi for all blocks in succession cient CS (q, 0), d. H. the coefficient with the lowest "Frequency", viewed as variable and by a umpte step of the approximation method mentioned above improves. Use further steps at this point would make little sense because of the marginal term of (1), like will become clearer below.

A unit 8 calculates the partial derivative for one of the sought coefficients CS (q, u) for the area term of F. The expression results for this derivative

and for the boundary term of F the expression

which is determined by a unit 9 .

Since it is an iterative approximation method to determine the zeros of the sum of (5) and the H times of (6), the zeroth approximation is used in (5) and (6) for IS (q, x) determined by unit 1 according to the above-mentioned method.

For the target boundary values IP (q, x) in the present case play an approximation of the signal you are looking for towards the point closest to an edge point x  but belongs to a neighboring block. Is an edge point x at the same time the edge point of a video image, is called Target margin value IS (p, x) used. The contribution of this edge point to (6) then becomes zero.

Via a line 12 a, a unit receives 12 information about the quantization characteristic used, that is also about the degree of quantization. The unit 12 makes µ smaller, the finer it has been quantized. With the finest quantization, the values IS (q, x) of the filtered signal should almost coincide with the decoded values I (q, x), since experience has shown that blocking is hardly detectable with the finest quantization. This plausible requirement for agreement is taken into account by the indicated dependency µ′s on the quantization characteristic, because the smaller u is, the more exactly IS (q, x) corresponds to I (q, x) in the minimum of F.

The current value of μ is multiplied by the expression ( 6 ) by a multiplier 10 and added to the term ( 5 ) by an adder 11 . The sum is fed to a control unit 7 via a line 11 a. The control unit 7 first calculates the second partial derivative of F according to the transformation coefficients CS (q, x), ie - except for factors - the expression

and determines from this the change δ (q, u) _{1 of} the u-th transformation coefficient according to the Newtonian approximation method. This change δ (q, u) is equal to the negative quotient of the signal on line 11 a and the expression ( 7 ). The signal values themselves are no longer included in the calculation of (7), but only values of the basic functions, the values of which are anyway only calculated once and then saved. In addition, the first term represents the norm of the u-th basis function, which becomes one with a suitable K in (3).

The control unit 7 passes the change δ (q, u) ₁ on to a unit 6 , which checks whether the total change, namely δ (q, u) ₀ + δ (q, u) ₁ , that the transformation coefficient C (q , u) has so far not exceeded a threshold D (q, u). In the example, this limit is the largest quantization error in terms of amount. The unit 6 can determine it from the data on the quantization characteristic, which is also supplied to it via line 12 a. The value of δ (q, u) receives _0, the unit 6 of the unit 1 through a line 1 b.

In the present example, D (q, u) is only for the first nineteen transform coefficients with the amount of maximum quantization error identical. For everyone else For the values of u, D (q, u) is set to zero, i.e. that is, this Coefficients are compared to the coefficients of I (q, x) no longer changed. The reason is that experience has shown that the quantization errors of the first Coefficients have a particularly strong impact on blocking, while the effect of the remaining coefficients is neglected is casual. The coefficients are counted after increasing amount of the vector u.

If the sum of all changes for a coefficient is less than or equal to D (q, u), as checked by the unit 6 , this coefficient is passed on to a unit 5 , which includes the current change δ (q, u) _{i of} the coefficient multiplies the associated basic function and passes it to an adder 4 . Further inputs of the Addie rers 4 are connected to the memory 3 , via which the adder 4 receives the values of the previous approximation for IS (p, x). The signal appears at the output of the adder 4

IS (p, x) _{i + 1} = IS (p, x) _i + δ (p, u) _{i + 1} Cos B (u, x), (8)

thus the (i + 1) th approximation for the filtered signal, which results from the (i + 1) th change of one of the transformation coefficients from the i th approximation for IS (p, x). If the sum of all changes is greater than D (q, u), this is reported to the control unit 7 .

The control unit 7 controls the entire process in such a way that initially only one iteration step for the coefficient with u = (0,0) is carried out per block in accordance with equation (8) and this iteration step is carried out for all blocks of a video image for which the sum of the Changes are within the allowable limits. When this run is finished, at least one further run for the next higher coefficient (s) is started under the same conditions. These are the coefficients with the vector u = (0.1) and with the vector u = (1.0). The next run then takes place for the vector u = (1,1) and so on up to the vector u = (6,6). Then δ (q, u) _{2 is} determined, that is, proceeded by one iteration step in the Newtonian method. The entire filtering is considered to have ended after three iterations for the Newtonian method.

The control unit 7 then causes the reading of the filtered signal from the memory 3 ; it is - after delivering an enable signal on a line 7 a - on a line 3 a are available.

The function of the circuit described can also be quite or partially by an appropriately programmed one Computers are taken over. Creating the or the In the present case, programs are part of the manual  Can the professional.

Because the indication of the signal flow direction in the figure and the characterization of the individual function blocks their signal-changing effects are for the expert Means that at the current state of simulation technology and the state of the higher programming languages (C, Fortran, Basic, DSPARC, DABL) equivalent to a flow chart are. Compare this as an example: Daisy manuals Systems Corporation "Simulation Compilation", August 1988 and "Daisy Behavioral Language", September 1988.

Claims (8)

1. Circuit arrangement for filtering a data signal transformed and quantized in data blocks, represented by the sample values I (q, x) of its pixels, the geometric position of which in a video image is determined by two-dimensional block coordinates q and relative coordinates x, or represented by transformation coefficients C ( q, u) with u as a two-dimensional variable in the transformation area, characterized by the following features:

• 1a) means ( 3 , 5 , 6 , 7 , 8 , 9 ) with which it is provided to determine a filtered video signal IS (q, x) by which the value of the functional under the secondary condition that the difference between a transformation coefficient CS (q, u) for the filtered video signal and the transformation coefficients C (q, u) of the video signal I (q, x) is smaller than or equal to a threshold D (q, u), whereby
  • a1) µ is an experimentally determined parameter,
  • a2) the sum in the second term for the functional F only runs over the relative coordinates x, which belong to the set Rq of the edge points of the block with the block coordinates q,
  • a3) IP (q, x) is a target boundary value for the boundary point x of the block with the block coordinates q,
• 1b) means ( 9 ) which are intended to determine the target boundary value IP (q, x) from an average value of the filtered video signal, averaged over the samples of pixels which are adjacent to the block with the block coordinates q.

2. Circuit arrangement according to claim 1, characterized in that the means ( 9 ) are provided for determining the target boundary value IP (q, x) from approximate values for the filtered video signal IS (q, x). 3. Circuit arrangement according to claim 2, characterized in that the means ( 4 , 5 ) are provided for that filtered video signal IS (q, x) to determine block by block as a linear combination of those basic functions which are used in the block-wise transformation of the video signal I (q , x) have been used and which are intended to make the functional F as small as possible with respect to the coefficients of the linear combination. 4. A circuit arrangement according to claim 3, characterized in that the means ( 7 ) are provided for making the functional F as small as possible for only one transformation coefficient and to carry out this process in a given sequence of the coefficients. 5. A circuit arrangement according to claim 4, characterized in that the means ( 7 , 8 , 9 ) are provided for applying Newtonian approximation methods in order to make the func tional F as small as possible. 6. Circuit arrangement according to claim 2, 3, 4 or 5, characterized in that the means ( 1 ) are intended to use the zero signal for the filtered video signal IS (q, x) the video signal I (q, x) , which is overlaid by a linear combination of the basic functions, the coefficients of which are determined as required. 7. Circuit arrangement according to one of the preceding claims, characterized in that the means ( 1 , 6 , 7 , 8 , 9 ) are intended to make F as small as possible under the secondary condition that the thresholds D (q, u) zero are when the amount of the vector u exceeds a predetermined value. 8. Circuit arrangement according to one of the preceding An claims, characterized, that programmed computers are provided for the radio tion of the specified funds in whole or in part to take.