KR100292348B1 - Initializing method according to the feature of a detected shape in block matching algorithm to apply genetic algorithm - Google Patents

Abstract

PURPOSE: A method for initiation according to characteristic blocks of an object extracting in a gene algorithm is provided to calculate an optimum motion value considering the characteristics of image blocks over a gene algorithm, thereby reducing a calculation amount and obtaining a precise motion value. CONSTITUTION: A method for initiation according to characteristic blocks of an object extracting in a gene algorithm includes the steps of dividing an image into predetermined blocks after extracting a predetermined object from a reference frame of the image, dividing per characteristics by relating the divided image blocks with the object by using the gene algorithm, and initiating a motion value by applying a motion value proper to the blocks divided per characteristics by the gene algorithm.

Description

Initializing method according to the feature of a detected shape in block matching algorithm to apply genetic algorithm in block matching method using genetic algorithm

The present invention relates to an image compression method, and more particularly, in a matching block method using a genetic algorithm, after dividing feature blocks in consideration of their association with a predetermined object extracted from a reference frame of an image, the separated feature blocks According to the initialization method to perform different initialization according to.

Image data compression technology uses intraframe coding to compress information using spatial redundancy present in an image, and temporal redundancy between these images when temporally consecutive images are present. ) Can be broadly divided into interframe coding for compressing information.

In particular, in video transmission, motion compensated coding using inter-frame coding is typically used as a method for reducing the amount of transmission data.

Motion compensation encoding refers to a method of transmitting a motion vector representing a displacement value of a moving object in a neighboring image frame in time and a prediction error according to the motion compensation.

As a method of measuring the amount of movement of an object, a block matching algorithm is used in a real-time video transmission system considering hardware.

The block matching algorithm includes a full search method, a three step search method using a fast algorithm, a one at a time search method, and a hierarchical block matching method using a hierarchical method. Algorithms (hierarchical BMA). Among them, the global search method has a regular data flow, which makes hardware implementation easier and shows good predictive performance compared to other methods, but has a large amount of calculation.

In addition, genetic algorithm is a search algorithm based on natural selection and genetic composition, which is an artificial system of adaptive phenomena of the natural world.

In order to implement the genetic algorithm, the object to be optimized must be represented by a set of parameters, that is, a meaningful bit string having genetic information. This work is an important part of the genetic algorithm, and it must be determined similarly to natural objects so that the bit strings can sufficiently represent the information we want.

The genetic information is determined by the change in the genotype in the transfer to the next generation, and the operator to choose the operator according to fitness as a transfer function for this. Unlike other search methods, genetic algorithms use internal genetic information, not direct phenomena, and deal with groups with individual genetic information so that changes in the population can be transformed into the desired optimization state.

Genetic algorithms are a kind of random search method, but unlike other random search methods, there are natural characteristics in addition to simple random properties, which are sets representing genetic information, that is, genotypes (DNAs or chromosomes) and their genetic changes. , And evolution. Therefore, genetic algorithms as search algorithms are not determined by random functions, but are expressed as natural ecosystem phenomena for sets representing genetic information. To this end, functions such as gene regeneration, breeding, and mutation are defined, and biochemical characteristics that change the genetic trait from one generation to another are evolved, ie, entropy decreases. In this way, the genetic algorithm minimizes the influence of local characteristics as a random function, and furthermore, it is possible to find global characteristics as natural phenomena determined by natural behavior.

In genetic algorithms, populations change their trait from one generation to the next by fitness, and the fitness plays a role in selecting only the appropriate genotype. This process requires genetic operators to change the genetic trait, and a simple genetic algorithm that shows good results in practical application problems consists of three operators: reproduction, crossover, and mutation. .

Regeneration is an operator that copies each genotype to the next generation according to their objective function, called the fit function. The objective function can be thought of as a measure of the optimization state according to the application problem. Copying genotypes by goodness of fit means that genotypes with high fitness are likely to survive the next generation.

After regeneration, hybridization is performed in two stages: first, the newly generated genotypes are randomly matched, then an arbitrary position of the genotype is selected, and the last bit of the genetic trait at that position matches the relative genotype. Change

Next is the variation that is caused by a given probability. It is an operator that converts any bits of a genetic trait into 0 for 1 and 1 for 0. These mutations are minor operators in genetic algorithms and can produce satisfactory optimization results by regeneration and crossover without mutation operators.

The present invention relates to a block matching method using a genetic algorithm that can reduce the amount of motion vector calculation by classifying macroblocks according to characteristics in association with an object extracted from a reference frame of an image and performing proper initialization according to the divided blocks. It is an object of the present invention to provide an initialization method according to the object extraction feature block.

1 is a view for explaining a block matching method using a genetic algorithm.

2 is a view for explaining an initialization method according to the present invention.

FIG. 3 is a diagram for describing a method of initializing by applying a genetic algorithm according to each type of FIG. 2.

4 is a diagram illustrating a search block.

In accordance with another aspect of the present invention, an initialization method in a block matching method using a genetic algorithm includes: an image segmentation process of extracting a predetermined object from a reference frame of an image and dividing the image into predetermined blocks; A feature classification process of classifying blocks of the divided image by each feature in association with the object; And an initialization process of initializing the motion value by applying a motion value suitable for the distinguished feature-specific blocks.

The distinguishing process for each feature is characterized by dividing the feature-specific block into blocks that are not related to the extracted object, blocks over the contour of the extracted object, and blocks within the extracted object. Blocks that are not related to the object apply the dynamic population control scheme (DPC), blocks across the contour of the extracted object apply the motion value of the weighted neighbor block, and blocks within the extracted object move the neighbor block. It is characterized by applying a value.

Hereinafter, with reference to the accompanying drawings will be described the present invention in more detail.

FIG. 1 is a diagram illustrating a block matching method using a genetic algorithm, in which an genetic algorithm is applied using correlation of motion values of adjacent blocks.

The motion value of the reference block is determined (S10). If the characteristics of the image are not considered, the initialization determines a randomly generated motion value as the initialization value, but since the video has a correlation with neighboring blocks, the average value obtained by averaging motion values of neighboring blocks using this correlation is used. Is the initial value. In this way, when the average value is initialized, the number of blocks to be searched can be reduced by reducing the number of iterations to which the genetic algorithm is applied, thereby reducing the calculation amount.

A fitness is calculated based on the initialized motion value (S12). As a method of calculating the goodness of fit, a mean square error (MSE) is used. In other words, the smaller the error value, the greater the goodness of fit.

A pair of populations having the smallest error value of the MSE used to calculate the goodness of fit, that is, a population having high goodness of fit, is selected (S14).

By performing the mating process to generate the next generation by combining or exchanging genes of the selected population (S16). Here, the mating process is performed through masking called UX (Uniform Crossover). That is, it generates a mask of the same length as the parent's chromosome length, where each bit is set to a random value. If one bit is 1, the offspring is one of two populations of chromosomes. Matches 0, and 0 matches one chromosome. The other descendant is determined using the inverted mask.

Since there may be missing traits in the population for statistical reasons, a mutant process that reverses the chromosome values is performed to compensate for this (S18). That is, the process of preventing the local optimal value from being found and finding the overall optimal value. to be.

Finally, repeating the steps S10 to S18 until the predetermined optimum value is reached, the strongest chromosome appears, and the value at this time becomes the overall optimal value.

2 is a view for explaining the initialization method according to the present invention, which shows the shape of a person extracted from a reference frame of an image, that is, an object.

Here, referring to the macroconvex divided in the reference frame, like the A1-B1 block, the block that does not belong to the extracted object (hereinafter referred to as A type), and the block including the contour of the extracted object, such as the A4-B2 block Blocks (hereinafter, referred to as type B) and A5-B3 blocks may be divided into blocks completely included in the extracted object (hereinafter referred to as type C).

Looking at the characteristics of each type, type A is a block that is difficult to predict the motion value, the initial value is the case that should refer to the motion value of the neighboring block, type B often have more than one motion value in the block Motion prediction using neighboring blocks has no effect. In addition, type C has a high probability of having the same movement value, and there are some slightly different movements. In particular, the human eye and mouth is a good example. This type of C is an example suitable for applying a genetic algorithm.

FIG. 3 is a diagram illustrating a method of initializing by applying a genetic algorithm according to each type of FIG. 2, wherein the population has four populations consisting of two parent chromosomes and two offspring chromosomes. Explain.

The feature of the macroblock is determined in association with the extracted object (S30).

As a result of the determination of S30, in the case of a block which is not related to the extracted object, for example, A type, a DPC (Dynamic Population Control Scheme) is applied. In other words, this method defines one motion value of the block at the same position in the previous frame and one motion value of the block available in the order of up, down, left, and right as the parent chromosome. In addition, two motion values are randomly selected from the blocks in the search region shown in FIG. 4 and replaced with the two chromosomes having the smallest fit in step 6. The DPC is performed once after the third generation, and a randomly selected value selects one from the search areas (1, 2, 3, 4) in FIG. 4 from the first and third search areas in the third generation, and 6 In the second generation, one from each of the search areas 2 and 4 is randomly selected. In this way, it selects 9th generation, 12th generation ... repeatedly. This order reflects the characteristic that the human eye is more sensitive to the vertical direction.

If it is a block associated with the object extracted in step S30, it is determined whether the block spans the contour portion (S34). If the block spans the contour portion, the weighted neighboring motion value is applied to the initialization (S36). That is, even if only a part of the object is located in the block, the motion of the object affects the entire block. However, since background motions cannot be ignored, two weights are given to the premise in this case, and one motion value is selected as a parent chromosome from type A and type C blocks among adjacent blocks, Cross to follow the characteristics of the parent chromosome to which they belong. In other words, instead of randomly masking the offspring, the offspring are generated by weighting them to better follow the C type.

As a result of the determination in S34, if the block does not span the contour portion, the neighboring motion value is applied to the initialization (S38). In this case, it is most preferable if a small motion can be found in the block in the divided object. This is because the overall motion is more likely to be the motion of the block itself. In this case, one motion value is selected as the parent chromosome from one motion value and one block at the same position of the previous frame, and then a small motion value can be found by an algorithm such as crossing and variation.

The present invention applies the characteristics of the image to the genetic algorithm in order to predict the optimal motion value. The most accurate method used for the motion value estimation is the global search method, but it requires too much calculation, and a simplified method is applied to solve this problem, but this has a problem of poor accuracy. In other words, the local optimal value is missing and the overall optimal value cannot be found. The nature of the genetic algorithm is that the overall optimal value can be found without inaccurate assumptions, but the computation is relatively large. To solve this problem, we proposed the block matching method in consideration of the characteristics of the image block in the genetic algorithm, which is less accurate and calculates the optimal motion value compared to the technique of applying the genetic algorithm based only on the motion value of the neighboring block. The motion value can be calculated.

Claims (3)

In the block matching method using a genetic algorithm, An image segmentation process of extracting a predetermined object from a reference frame of the image and dividing the image into predetermined blocks; A feature classification process of classifying the blocks of the divided image by the features by associating the object with the object using the genetic algorithm; And And an initialization process of initializing a motion value by applying a motion value suitable for the classified feature blocks using the genetic algorithm. The method of claim 1, wherein the classification process for each feature is performed. An initializing method characterized by dividing blocks by features into blocks not related to the extracted object, blocks across the contour of the extracted object, and blocks within the extracted object. The method of claim 1 or 2, wherein the initialization process is performed. Blocks not related to the extracted object apply a dynamic population control scheme (DPC), blocks across the contour of the extracted object apply the motion value of the weighted neighboring block, and blocks within the extracted object Initialization method characterized in that the motion value of the block is applied.

Images