JPH07123276A - Digital compression coding method for picture signal - Google Patents

Abstract

PURPOSE:To improve the quality of an approximated picture and to reduce the processing time by using an optimizing method so as to optimize and estimate a parameter used for description of a nonlinear reduction map function. CONSTITUTION:An input picture signal 1 is given to a function parameter estimate section 2. The estimate section 2 optimizes and estimates a map function parameter 4 used to describe a nonlinear reduction map function 3 by using a map function parameter optimization section 7. Thus, an arithmetic quantity required for estimating a parameter is considerably reduced and the estimate of the parameter is made efficient. Furthermore, the optimization section 7 preferably has an initial gene generation stage in which a parameter describing a nonlinear reduction map function is mapped into a symbol string called a gene and has a gene evolution stage in which a gene control such as crossover, reversing and mutation is applied repetitively to a mass of generated initial genes thereby replacing a gene with a low adaptability with a newly generated excellent succeeding gene for sorting.

Description

Detailed Description of the Invention

[0001]

The present invention relates to image communication, image recording,
The present invention relates to a digital compression encoding method of an image signal applicable to fields such as computer graphics.

[0002]

2. Description of the Related Art Conventionally, an image signal is subjected to discrete cosine transform (
_{Discrete Cosine Transfrom} ) has been proposed as a digital compression encoding method for image signals, in which an image signal is digitally compression encoded.

According to this digital compression coding method for image signals, most of the image signal components are concentrated in the low-order coefficients due to the discrete cosine transform of the image signals, so that the probability distribution of the image signals is as low as the lowest-order discrete cosine coefficient. Therefore, the image signal can be compressed more easily than the original image signal having a flat probability distribution.

Further, conventionally, the image signal is digitally combined by performing a discrete cosine transform of the image signal and a process of roughly quantizing (or truncating) a wide-area component of the image signal having low human visual sensitivity. A digital compression encoding method of an image signal called compression encoding has also been proposed.

According to such a digital compression encoding method of an image signal, the image signal is further compressed as compared with the case where the image signal is digital compression encoded only by performing the discrete cosine transform of the image signal. You can

However, since the compression limit of the image signal by the discrete cosine transform of the image signal is defined by the rate distortion theory, there is a certain limit in digitally encoding the image signal at a high compression rate. It had drawbacks.

Therefore, conventionally, paying attention to the self-similarity of the image signal, the image signal is approximately represented by a reduction mapping function system that iteratively repeats the reduction mapping, and
There has been proposed a digital compression coding method of an image signal in which a parameter describing a reduction mapping function at that time is coded.

According to such a digital compression encoding method of an image signal, the image signal can be digitally encoded at a higher compression rate than in the case of performing the digital compression encoding with the discrete cosine transform described above. it can.

Here, the mapping function system in such a digital compression coding method for image signals will be described. The mapping function system includes a linear mapping and a non-linear mapping, and the one-dimensional linear mapping in the linear mapping is performed. The mapping (linear transformation) is expressed by x _{k + 1} = ax _k + b (k = 1, 2, ..., K) (1). Here, if a <1, that is, the reduction map, there exists an invariant set (attractor) that does not depend on the initial set (initial image). Therefore, if the reduction map function is repeatedly executed on any initial image, , The image signal can be reproduced.

The two-dimensional linear transformation in the linear mapping as the above mapping function system is called an affine transformation,

[Equation 1] ……………… (2) is used to represent a combination of image rotation, reduction or enlargement, and translation.

Further, the non-linear mapping (non-linear conversion) as a mapping function system is a one-dimensional non-linear mapping (non-linear conversion).
Take a simple example of

[Equation 2] ............ It is expressed by (3), and it is possible to describe any mapping that cannot be expressed by the above-mentioned linear mapping (linear transformation).

Therefore, in the digital compression coding method of the image signal, the image signal is approximately expressed by the reduced mapping function system, and the parameters describing the reduced mapping function at that time are coded. In the case of a linear mapping (linear transformation) (affine transformation), an image can only be represented by rotation, reduction or enlargement, and parallel movement, but when the reduction mapping is a non-linear mapping (non-linear transformation), it is
If it is a non-dimensional mapping (non-linear transformation) of dimension,
Various non-linear deformations such as rotation, contraction or enlargement, translation, translation, and distortion can be expressed.

From the above, the digital compression code of the image signal is such that the image signal is approximated by the non-linear reduction mapping system, not by the linear reduction mapping system, and the parameters describing the non-linear mapping at that time are encoded. The encoding method is more useful than the other digital compression encoding methods for image signals described above.

[0014]

However, the above-mentioned conventional image in which the image signal is approximately represented by the non-linear reduction mapping function system and the parameters for describing the non-linear reduction mapping function at that time are encoded. In the case of the digital compression encoding method of the signal, as shown in FIG. 4, the input image signal 1 is input to the function parameter estimation unit 2, and the function parameter estimation unit 2 accurately inputs the input image signal 1. An approximate non-linear reduction map function parameter 4 is estimated as a parameter describing the non-linear reduction mapping function by using a mapping function 3 to which a predetermined non-linear transformation can be applied, and the function is estimated. The non-linear reduction mapping function parameter 4 designated by the parameter estimation unit 2 is encoded into the code data 6 by the encoding unit 5, and the code data 6 is transmitted or stored. However, in estimating the non-linear reduction mapping function parameter 4, and thus the parameters describing the non-linear reduction mapping function, the possible values of the function parameters are limited in advance, and The reduction map is executed for all combinations (brute force search) to find the combination of parameters that minimizes the distance between the approximate image and the input image by the reduction map.

For this reason, the above-mentioned conventional digital compression code of an image signal is obtained by approximating the image signal by a non-linear reduction mapping function system and encoding the parameters describing the non-linear reduction mapping function at that time. In the case of the conversion method, there is a drawback that the value that the function parameter can take and the number of parameters increase.

Further, there is a drawback that a large amount of calculation is required for estimating the above-mentioned non-linear mapping function parameter.

For this reason, it is conceivable to limit the values of the possible function parameters to the range of the practical calculation amount. In this case, however, by searching only this limited solution candidate set, It has a drawback that it is difficult to obtain a true optimal solution.

[0018]

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to express an image signal approximately by a non-linear reduction mapping function system, and to encode parameters for describing the non-linear reduction mapping function at that time. In the digital compression coding method, the parameters for describing the non-linear reduction mapping function are optimized and optimized to improve the quality of the approximate image and reduce the processing time.

[0019]

According to a first aspect of the present invention, there is provided a digital compression encoding method for an image signal, wherein the image signal is approximately represented by a non-linear reduction mapping function system, and a parameter for describing the non-linear reduction mapping function at that time is described. In a digital compression coding method of an image signal in which a parameter is coded, parameters for describing the non-linear reduction mapping function are optimized and estimated using an optimization method.

According to the digital compression encoding method for an image signal according to the second invention of the present application, in the digital compression encoding method for an image signal according to the first invention of the present application,
A genetic algorithm is applied to an optimization method of parameter estimation that describes a nonlinear reduction mapping function. In this genetic algorithm, for the initial gene generation step of mapping the parameters describing the non-linear reduction mapping function to a symbol string called a gene, and for the population of initial genes generated in the initial gene generation step, It has a gene evolution stage in which gene manipulations such as crossover, inversion, and mutation are repeatedly performed, and low fitness genes are replaced by newly generated excellent offspring genes.

Here, the genetic algorithm will be described. It is a model that applies natural selection of the living world and a genetic phenomenon, and the composition of the population is determined by the set of individuals (solution candidates in the search problem) in the external environment ( In order to adapt to the evaluation function in the search problem, (1) individuals with higher fitness have higher survival probabilities (natural selection) (2) rules that generate new individuals based on old individuals (genetic phenomenon) Based on the above-mentioned rules, the update is performed for each generation. According to such a genetic algorithm, the total number of individuals is constant because it is based on the rules described above. When such a restriction (selection pressure) is applied, a competitive relationship occurs within the population, and a population more adapted to the environment succeeds in breeding.

Further, the above-mentioned genetic algorithm will be exemplified here.

Now, an individual group X = {x ¹ , ... ......, consisting of M individuals each represented by a gene (symbol) of length n
x ^M }.

Procedure 1: The population X (0) = {x ¹ (0), ..., X ^M (0)} of the 0th generation (the number of generations t = 0) is
Randomly created, the fitness v (x ⁱ (0)) is calculated for each x ⁱ (0) in X (0).

Step 2: For each x ⁱ (0) in X (t), the selection probability shown in the following equation (4)

[Equation 3] To calculate.

[Equation 4] ……………… (4)

Step 3: 2 individuals x from X (t)
ⁱ (t) and x ^j (t) are selected probabilities

[Equation 5] as well as

[Equation 6] Select each according to. (1) Crossover is applied to two individuals with a certain probability P _e , and one is randomly selected from the two descendants generated thereby, which is ^defined as x ^k (t). (2) x ^k (t), applied with a probability P _i that a reversal is x ^k which has been subjected to the reverse (t) as a new x ^k (t). (3) The probability P _m that there is a mutation in each element of x ^K (t)
And apply the applied x ^k (t) again to x ^k (t)
And

Step 4: From X (t), one individual is selected according to a certain rule, and it is replaced with x ^k (t).

Step 5: Calculate the fitness v (x ^k (t)).

Step 6: End condition (for example, generation number t = 10)
00) is satisfied, the generation t is set as t ← t + 1,
Repeat steps 2 to 5.

Here, the genetic operators, crossover, inversion, and mutation in the above-mentioned algorithm will be briefly described.

Crossover: Explaining the simplest one-point crossover (one-point crossover), two individuals (n = 10) represented by the parent x ⁱ = 0010111011 x ^j = 0111100111 are selected. If this is the case, one of the break intersection points (crossover points) of the bit string is selected with a probability P _e of 1 / (n−1). Now, if the break crossing point is chosen as the fourth, the parent's 4
By exchanging up to the 5th bit and after the 5th bit, two descendants represented by x ^p = 0010 | 100111 x ^q = 0111 | 111011 are created.

Then, one of the descendants x ^p and x ^q is selected, and it is set as x ^k (t).

Inversion: In the simplest case of inversion, two discontinuities are randomly selected from the gene bit sequence of x ^k (t), and the bit sequence between the discontinuities is changed. If x ^k (t) = 0111111011 and the cuts are selected at the third and eighth positions as represented by 011 | 11110 | 11, then 011 | 01111 | 11 is set.

Mutation: Producing offspring from two parents by crossover produces only a limited range of offspring compared to the size of the space to be searched. Therefore, descendant x
Inverting {0, 1} of each bit of ^k (t) serves to expand the search range.

[0034]

[Operation] According to the digital compression encoding method of an image signal according to the present invention, a parameter for describing a non-linear reduction mapping function is used.
Since the estimation of the data is optimized by the optimization method,
The amount of calculation required for parameter estimation can be greatly reduced as compared with the conventional digital compression coding method for image signals in which the parameters are not optimized and estimated by the optimization method.

Further, since the genetic algorithm is used as the parameter optimizing means, a selection pressure of a constant number of individuals is applied to the group of genes (combination of mapping function parameters) to cause a competitive relationship within the group. Occurs, and parameter optimization is performed according to the laws of organism evolution. In particular,
Individuals with high adaptability to the environment, that is, combinations of parameters with a small distance between the approximate image and the original image obtained by the non-linear reduced mapping function system succeeded in surviving, repeated evolution with generations, and finally It is possible to find the combination of function parameters with the highest fitness.

Further, by applying inversion or mutation to a gene, it is possible not only to search only a predetermined candidate as in the conventional method but also to search a wider arbitrary solution candidate space, It is possible to avoid premature convergence to a local solution, and thus a non-linear reduction map function parameter.
Can be obtained as a global optimum solution, and thus the quality of the approximate image can be improved.

[0037]

Embodiment 1 Next, with reference to FIG. 1, a principle embodiment of a digital compression encoding method of an image signal according to the present invention will be described.

In FIG. 1, parts corresponding to those in FIG. 4 are designated by the same reference numerals.

In the principle embodiment of the method for digitally compressing and encoding an image signal according to the present invention shown in FIG. 1, as in the case of the conventional method for digitally compressing and encoding an image signal described above with reference to FIG. Signal 1 is used as a function parameter estimator 2
To the function parameter estimator 2 for accurately expressing the input image signal 1 accurately.
Is estimated as a parameter describing a non-linear reduction mapping function using a mapping function 3 designated in advance.

However, in the principle embodiment of the digital compression encoding method of the image signal according to the present invention, the estimation of the function parameter 4 in the function parameter estimation unit 2 is performed as follows.
The parameter optimization unit 7 is used for optimization.

Further, the function parameter 4 output from the function parameter estimation unit 2 and estimated by the parameter optimization unit 7 is estimated in the encoding unit 5 by the encoding unit 5. , And the code data 6 is transmitted or stored.

[0042]

[Embodiment 3] Next, a concrete embodiment of a digital compression encoding method of an image signal according to the present invention will be described with reference to FIG.

In FIG. 2, parts corresponding to those in FIG. 1 are designated by the same reference numerals.

In the concrete embodiment of the digital compression encoding method for an image signal according to the present invention shown in FIG. 2, the input image signal is input in the same manner as in the digital compression encoding method for an image signal according to the present invention shown in FIG. 1 is a function parameter estimation unit 2
To the function parameter estimator 2 for accurately expressing the input image signal 1 accurately.
Is estimated by using the mapping function 3 designated in advance and using the parameter optimizing unit 7.

In this case, in the parameter optimization unit 7, in order to apply the above-described gene algorithm, the initial gene generation unit 9 is used to call a group (group) 8 of arbitrary initial parameters as a gene. Is performed to generate and output the initial gene group (group) 10, and then the initial gene group (group) 10 output from the initial gene generation unit 9 is converted into the gene evolution unit 11. , The gene evolution unit 11 repeatedly performs genetic operations such as crossover, inversion, and mutation, and replaces the low-fitness gene with a newly generated excellent offspring gene to eliminate the function parameter. -Optimize and estimate parameter 4.

The optimized non-linear mapping function parameter 4 output from the function parameter estimation unit 2 is coded into the code data 6 in the coding unit 5, and the code data is coded. 6 is transmitted or stored.

The code data 6 transmitted or accumulated according to the above-described embodiment of the digital compression encoding method of the image signal according to the present invention is processed by the decoding unit 12 in a non-linear manner as shown in FIG. The non-linear mapping function parameter 13 is decoded, and the non-linear mapping function parameter 13 is input to the image reproducing unit 14, so that the non-linear mapping function parameter is used in the image reproducing unit 11 to reproduce the reproduced image signal 15 Is what you can get.

[0048]

EFFECTS OF THE INVENTION According to the digital compression coding method for image signals of the present invention, (1) Parameter estimation is performed because the nonlinear mapping function parameter is estimated by using the parameter optimization method. Therefore, the processing time can be reduced compared to the conventional brute force search. (2) By applying the genetic algorithm to the optimization of the mapping function parameters, the conventional method can be used. It is possible not only to search for a predetermined candidate but also to search a wider range of solution candidate spaces, avoiding early convergence to a local solution, and finding a global optimum solution. (3) Since the mapping function parameter is obtained as a global optimum solution, the effect of improving the approximation accuracy of the image and improving the quality of the reproduced image can be obtained.

[Brief description of drawings]

FIG. 1 is a system diagram showing a principle embodiment of a digital compression encoding method of an image signal according to the present invention.

FIG. 2 is a system diagram showing a specific example of a digital compression encoding method of an image signal according to the present invention.

FIG. 3 is a diagram showing a system for decoding data encoded by the digital compression encoding method for image signals according to the present invention.

FIG. 4 is a diagram showing a conventional digital compression encoding method for image signals.

[Explanation of symbols]

 1 input image 2 function parameter estimation unit 3 mapping function 4 mapping function parameter 5 coding unit 6 code data 7 mapping function parameter optimization unit 8 initial parameter group 9 initial gene generation unit 10 initial Gene group 11 Gene evolution unit 12 Decoding unit 13 Mapping function parameter 14 Image reproduction unit 15 Reconstructed image

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Hiroshi Kodera 1-1-6 Uchisaiwaicho, Chiyoda-ku, Tokyo Nihon Telegraph and Telephone Corporation (72) Tomio Kishimoto 1-1-6 Uchisaiwaicho, Chiyoda-ku, Tokyo No. Japan Telegraph and Telephone Corporation

Claims (2)

[Claims] 1. A digital compression encoding method for an image signal, wherein an image signal is approximately represented by a non-linear reduction mapping function system, and a parameter describing a non-linear reduction mapping function at that time is encoded. A digital compression coding method for an image signal, characterized by optimizing and estimating a parameter describing a reduction mapping function using an optimization method. 2. The digital compression coding method for an image signal according to claim 1, wherein the optimization method maps a parameter describing the non-linear reduction mapping function to a symbol string called a gene. Gene manipulations such as crossover, inversion, and mutation are repeatedly performed on the generation stage and the group of initial genes generated in the initial gene generation stage, and genes with low fitness are transformed into newly generated superior child genes. A method for digital compression encoding of an image signal, characterized in that it has a gene evolution step of replacing and selecting.