JP2925055B2 - Image coding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image encoding apparatus for encoding an input image signal, and more particularly to an image encoding apparatus using subband division.

[0002]

2. Description of the Related Art As an image encoding apparatus for encoding an input image signal, those employing various systems are known. One of them is an image encoding apparatus using subband division. There is.

In an image coding apparatus using this sub-band division, an input image signal is decomposed into a plurality of sub-band signals having different frequency bands, and appropriate bits are allocated to each sub-band. A high coding efficiency is obtained by expressing the band signal with a smaller number of bits than the number of bits of the original signal.

The input image is divided into sub-bands of a two-dimensional frequency band by performing the sub-band division on the signals in the vertical and horizontal directions or by connecting in multiple stages. That is, the frequency of the image is as shown in FIG.
Can be expressed in two dimensions as follows. This two-dimensional frequency domain is divided into a plurality of frequency domains, for example, as shown in FIG. The divided frequency domain signal becomes a subband signal.

In encoding the sub-band signal, there is known a method of allocating appropriate bits to each sub-band and performing quantization to obtain high encoding efficiency (hereinafter, referred to as conventional method 1). (JW Wood
s, S. D. O'Neil, “Subband Cod
ing of Images ”, IEEE tran
s. , Acoustic. , Speech & Signa
l Process. , ASSP-34, No. 5, O
ct. 1986, pp. 1278-1288).

In general, assuming that the probability density function of each subband signal distribution is the same, if the variance of each subband signal is large, the number of bits allocated to that subband is increased and the number of bits allocated to the subband is small. Sometimes, the average distortion can be minimized by reducing the number of bits allocated to the subband.

In the above-mentioned conventional method 1, the probability density function of the previous difference signal of each subband signal is assumed to be a Laplace distribution. Further, the variance of the difference signal of each sub-band signal is measured, and the bit allocation to each sub-band is determined by the following equation.

[0008]

(Equation 3) Where B _i is the number of bits allocated to each subband i, σ _i is the positive square root of the variance of the signal of subband i,
ξ is a parameter that defines the image quality and the total code amount.

[0009] However, in the above-mentioned conventional method 1,
There are problems as described below.

[Problem 1] In the conventional method 1, since one probability density function is applied to all signals, a mismatch at the time of quantization occurs, and optimal coding is not always performed. There is a disadvantage that there is no.

In general, not all subband signals have the same distribution. However, in the conventional method 1, since bits are allocated using only the variance value, necessary bits may not be allocated in some cases.

An example of a distribution having the same variance but a different shape will be described below.

For example, as shown in FIGS. 5A and 5B, there are distributions having the same variance σ ² but different shapes. In the case of the signal distribution shown in the figure, even if the variance values are the same, the average distortion is smaller when more bits are allocated to the distribution (a) than in the distribution (b).

[0014] In [Problem 2] conventional method 1, in the formula (1) for obtaining the number of bits B _i assigned to each sub-band i mentioned above, when obtaining the parameters xi], the negative number of bits B _i To avoid this, we use an iterative method.
That is, when the bit B _i is negative, B _i = 0 and
I am doing the calculations again. And all the bits Bi
Is repeated until is equal to or greater than 0.

As described above, in the conventional method 1, the number of bits B
_Since iterative calculation is performed to obtain _i , the processing time cannot be constant. If the processing time cannot be kept constant, the worst case must be estimated and the processing time must be considered, resulting in a longer encoding processing time.

In order to solve the above problem 1, there has been proposed a method of performing bit allocation in consideration of a distribution shape (hereinafter, referred to as a conventional method 2) (YQ Zhang, S. Za).
far, "Motion-Compensated W
avelet Transform Coding f
or Color Video Compression
n ", Proc. SPIE VCIP'91, Vol.
1605, pp. 301-316, Nov. , 1991
reference).

In the conventional system 2,

(Equation 4) The bit distribution is performed by obtaining ξ. Where α _i
Is asked when the probability density function for each sub-band signal distribution put the _{p i (x), α i} = [∫ [p _i (x)] ^1/3 dx] ³ × 2 ^2m ··· ⁽³⁾ Parameter. Further, 2 ^2m is a coefficient for weighting the low band. Here, m is a parameter given to each subband as shown in FIG. Parameter m
Indicates that the sub-band is a band that is sub-sampled vertically and horizontally by １ ／ m each.

The parameter α _i in the equation (2) is a number corresponding to the variance sensuously, but the difference from the variance is that the distribution shape is taken into consideration.

As described above, in the conventional system 2, by allocating bits using the equation (3), it becomes possible to allocate bits according to the probability density function of each subband signal distribution.

However, the conventional method 2 still has a problem 2 in the conventional method 1, that is, a problem of requiring an iterative calculation and increasing the processing amount.

[0021]

SUMMARY OF THE INVENTION It is therefore an object of the present invention to reduce the number of repetitive calculations so that bit allocation according to the subband signal distribution can be performed with a small amount of processing.

[0022]

According to the present invention, there is provided a subband dividing step of dividing an input image signal into a plurality of subband signals having mutually different frequency bands, and a plurality of subband signals obtained from the subband dividing step. A bit allocation step of determining the number of bits to be allocated at the time of quantization in accordance with the following equation; and a quantization step of quantizing each subband signal from the subband division step according to the number of allocated bits determined by the bit allocation step. And wherein the number of bits Bi assigned to each subband _i is determined by the following calculation in the bit assignment step:

(Equation 5) Where: a _i = [∫ [p _i (x)] ^1/3 dx] ³ × 22 ^m p _i (x): probability density function of each subband signal distribution 2 ^2m : coefficient for weighting low band : when obtaining the parameter the parameter ξ defining the picture quality and the total amount of codes, the total number of bits B, and the number of each sub-band signals i and n _i, and a _j are arranged from the larger to sort the a _i At the same time, _ni is arranged to correspond to the rearrangement of a _i to be n _j, and the calculation of the following expression is repeated while increasing the value of k in order,

(Equation 6) The calculation is stopped when d (k)> A _j, and the parameter ξ is obtained from ξ = d (k−1) at this time.

[0023]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The features of the present invention will be specifically described based on embodiments with reference to the drawings. First, the basic concept of the present invention will be described.

In the present invention, the above equation (2), that is, considering the distribution shape at the time of bit allocation, ie,

(Equation 7) Is performed.

At this time, in order to solve the problem that the amount of calculation is extremely large because it is necessary to perform numerical integration in order to obtain the parameter α _i , a feature amount previously obtained from an input signal, for example, Standard deviation and absolute mean,
By preparing a look-up table for obtaining the relationship with the parameter α _i and using this look-up table, α _i can be obtained without using a numerical calculation with a large processing amount.

Here, a generalized Gaussian distribution G _{a, b} (x) is applied as a probability density function of each subband signal distribution. The generalized Gaussian distribution G _ab (x) is

(Equation 8) However,

(Equation 9) Here, Γ () is a gamma function.

In this generalized Gaussian distribution, the parameter a
And b, G _{a, b} (x) can be determined.

By using such a general-purpose distribution function, a probability density function can be easily specified by obtaining only two parameters.

In order to obtain the two parameters a and b, the relationship between some independent two characteristic amounts of the subband signal (for example, the standard deviation of the subband signal and the average of the absolute value) and the parameter is obtained. It should be left.

If the relationship between the two parameters a and b of the generalized Gaussian distribution and the parameter α _{i in} equation (3) is obtained, the parameter α _i can be finally obtained.

However, since it is necessary to find α _i after all, it is not necessary to find the parameters of the generalized Gaussian distribution. That is, a relational expression for directly obtaining the parameter α _i from the feature amount of the subband signal may be obtained.

Therefore, here, a look-up table that describes the relationship between some independent two feature values of the subband signal and α _i is obtained. As the two feature values, for example, it is conceivable to employ a standard deviation of a subband signal and an average of absolute values.

Since the parameter b of the generalized Gaussian distribution is a standard deviation, it can be obtained immediately. Therefore, in the following calculation, it is sufficient to consider all signals in a state of being divided by the standard deviation and normalized. That is, b = 1.
It is sufficient to finally multiply b by setting it to 0.

After all, a look-up table for converting the average of the absolute values of the sub-band signals into a normalized value of α _i in equation (3) may be created according to the following procedure.

First, β is defined as β = [∫ [G _{a, 1.0} (x)] ^1/3 dx] ³ . This β is normalized by removing the weighting coefficient of the parameter α _{i in} equation (3). Further, it is assumed that X = ∫ | x | × G _{a, 1.0} (x) dx. X is the average of the absolute values of the normalized generalized Gaussian distribution.

By plotting X and β while changing the parameter a, for example, a relationship as shown in the graph of FIG. 1 is obtained. Next, as shown in FIG. 5, representative points (indicated by crosses) in the graph of FIG. 4 are taken. Further, a look-up table as shown in Table 1 is created.

[0037]

[Table 1] This look-up table describes the relationship between X, which is the average of the absolute values of the signals normalized by the standard deviation, and β, which is obtained by removing the weighting of α _i and normalizing it.

Using this look-up table, the parameter α _i is obtained as follows.

First, the standard deviation σ and the average absolute value Y of the input signal are obtained. Next, Y / σ is compared with the look-up table in Table 1 to find the applicable β. If Y / σ is not a representative value of X, β is obtained by an interpolation process of extrapolation. Further, according to the following equation, multiply β by a standard deviation,
Α _i can be obtained by weighting.

Α _i = β · σ ² × 2 ^{2m In the} present invention, the equation (2) is used as described above.

(Equation 10) In, bit allocation is performed by obtaining a parameter ξ that defines the image quality and the total code amount. However, equation (2) is
has a value only when a _i ≧ ξ, otherwise B _i =
Set to 0.

Now, assume that the total number of bits required is B and the signal i
, And the maximum value of the total number of classes, i.e., i, is N. From equation (2), the total number of bits B is

[Equation 11] Der Ru. Substituting Bi in equation (2) into Bi in equation (4)
And

(Equation 12) Becomes If you transform this,

(Equation 13) Becomes

Here, equation (2) is an equation that has a value only when a _i ≧ ξ, and B _i = 0 otherwise.
If equation (2) is calculated as it is, ξ such that a _i <ξ may be a solution. To avoid this, the following method is adopted.

If a _i <ξ, B _i = 0 in equation (2)
This means that B _i that satisfies a _i <ξ does not need to be calculated because it is originally 0. In other words, if we calculated from the larger of a _i, somewhere in a _i
<Ξ can be found. Thereafter, it is sufficient to set B _i = 0, so that there is no need for calculation.

Therefore, in the present invention, when obtaining ξ, _ai is sorted and calculated.

It is assumed that A _i is sorted and arranged in descending order of A _i .

[0046]

[Equation 14] And The equation where k = N in this equation is the same as the equation (6).

The ξ to be obtained is any of d (k) (where k = 1, 2,..., N). If k is increased sequentially in equation (7), there is a point where A _j <d (k) somewhere. At the time when A _j <d (k), a case where a _i <ξ has already been added, so this case needs to be removed. That is, since the solution calculated immediately before is a solution, ξ = d (k−1). In addition,
A _j <d (k) may not be satisfied until k = N, but in this case, ξ = d (N) is the solution.

The sort processing of _ai described above can be realized by a sort circuit constituted by hardware or software processing. In the present invention, by performing this sort processing, the conventionally required repetition processing becomes unnecessary.

Here, the amount of calculation and the calculation time in the conventional example and the present invention will be compared.

It is assumed that the number of subbands or classes is N. In the conventional example, first,

(Equation 15) Is calculated. If there is a sub-band (or class) where B _i <0 as a result, the calculation of equation (8) is performed again, excluding α _i , assuming that the bit allocation is B _i = 0. This is performed until all B _i ≧ 0. Therefore, it is necessary to perform the calculation of Expression (8) once or more.

On the other hand, the calculation of equation (7) shows that k is N
Even if the calculation is performed up to, the calculation amount is equivalent to the calculation of Expression (8).

Now, assuming that the calculation amount of equation (8) is P, the calculation amount of the sorting circuit is Q, the calculation time of equation (9) is p, and the calculation time of the sorting circuit is q, the calculation amount of the conventional example is mP (However,
While m is an integer of 1 or more, the calculation amount of this method is (P−δ) + Q (where δ is a number from 0 to less than P). Further, while the calculation time of the conventional example is mp (where m is an integer of 1 or more), the calculation time of this method is (p−
δ) + q (δ is a number from 0 to less than p).

In the conventional method, since m is not determined in advance, the calculation amount and the calculation time cannot be constant. On the other hand, in this method, the amount of calculation and the calculation time can be reduced to a certain value or less by using a sort circuit. In particular, when the number of subbands (or classes) is small, the sorting load is reduced, which is particularly effective.

FIG. 3 is a block diagram showing an embodiment of an image coding apparatus to which the present invention is applied.

Referring to FIG. 1, reference numeral 1 denotes a sub-band division unit for dividing an input image signal into a plurality of sub-band signals having mutually different frequency bands, and 2 denotes a sub-band signal obtained from the sub-band division unit 1. The bit allocation unit 3 that determines the number of bits to be allocated at the time of quantization, quantizes the sub-band signal divided by the sub-band division unit 1 with the number of bits determined by the bit allocation unit 2, and generates a quantization index. This is a quantization unit for outputting.

The bit allocation unit 2 includes a statistic detection unit 2a
, A lookup table 2b, a calculation unit 2c, and a calculation unit 2d. The statistic detection unit 2a calculates a standard deviation σ and an average absolute value Y of the signal. As shown in Table 1, the look-up table 2b is obtained by normalizing the average of the absolute values of the signals by the standard deviation X and the parameter α _i
The relationship of β obtained by removing the weights and normalizing the weights is described. The operation unit 2c performs an operation for obtaining the parameter α _i , and the operation unit 2d performs an operation for obtaining the assigned bit number B _i .

Hereinafter, the operation of the image encoding apparatus shown in FIG. 1 will be described.

The input image signal is divided into a plurality of sub-band signals by the sub-band dividing section 1. Each sub-band signal from the sub-band division unit 1 is supplied to a statistic detection unit 2a of the bit allocation unit 2, and a standard deviation σ and an average absolute value Y of the signal are obtained. Next, Y / σ is compared with the look-up table in Table 1 to find the applicable β. Further, the following equation is calculated in the calculation unit 2c, and the parameter α _i is obtained.

Α _i = β · σ ² × 2 ^2m Then, the following equation is calculated in the arithmetic unit 2d, and the number of allocated bits B _i is obtained.

[0060]

(Equation 16) Based on the number of bits thus obtained, the quantization unit 3 performs quantization for each subband.

In the above, α _i obtained by equation (3)
, But the fraction of the subband signal may be used instead of α _i .

[0062]

As described above, according to the present invention, since α _i is sorted in advance and 求 め る is calculated in a range where the number of bits is 0 or more, useless calculation is performed. And the calculation time can be reduced.

[Brief description of the drawings]

FIG. 1 is a graph of X defined by X = ∫ | x | × G _{a, 1.0} (x) dx and β = β defined by [∫ [G _{a, 1.0} (x)] ^1/3 dx] ³ It is a graph which shows a relationship.

FIG. 2 is an explanatory diagram showing a procedure for creating a lookup table.

FIG. 3 is a block diagram illustrating an embodiment of an image encoding device to which the present invention is applied.

FIG. 4 is an explanatory diagram showing subband division.

FIG. 5 is an explanatory diagram showing an example in which the distribution is different with the same variance.

FIG. 6 is an explanatory diagram showing parameters given to each subband.

[Explanation of symbols]

1 ... subband splitting unit, 2 ... bit allocation unit, 2a ... statistic detector, 2b ... lookup table, 2c ... alpha _i arithmetic unit, 2d ... B _i arithmetic unit, 3 ... quantizer

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) H04N 1/41-1/419 H04N 7/24-7/68

Claims (1)

(57) [Claims] 1. A sub-band division unit for dividing an input image signal into a plurality of sub-band signals having mutually different frequency bands.
A step, the bit allocation step of determining the number of allocated bits at the time of quantization in accordance with a plurality of sub-band signals obtained from the subband dividing step, the bit allocation to each subband signal from the subband dividing step an image encoding method comprising the quantization step of quantizing in accordance with the number of allocated bits determined by the process, the bit allocation step smell
The number of bits B _i allocated to each subband _i is determined by the following calculation. Where: a _i = [∫ [p _i (x)] ^1/3 dx] ³ × 22 ^m p _i (x): probability density function of each subband signal distribution 2 ^2m : coefficient for weighting low band : when obtaining the parameter the parameter ξ defining the picture quality and the total amount of codes, the total number of bits B, and the number of each sub-band signals i and n _i, and a _j are arranged from the larger to sort the a _i And to sort _ai
Correspondingly, n _i is arranged to be n _j, and the calculation of the following equation is repeated while increasing the value of k in order. d (k)> Stop calculated at time point when A _j, image marks, characterized in that determining the parameters xi] from at this point ξ = d (k-1)
Encoding method .