KR20060054976A - Method of decoding encoded information of smell information or feeling of smell transferred from video image - Google Patents

Abstract

The present invention provides a method of encoding an expression factor for a feeling of smell of an odor information or a video image and then reducing the dimension by PCA and restoring the input encoding information based on the representative smell by FCMA. That is, the present invention provides a method of digitally decoding a smell of an odor information or an image image, the method comprising: performing encoding according to a factor expression of a representative odor and obtaining and storing a center value of the representative odor by clustering; Receiving the odor information encoded and transmitted and mapping the pattern of the data to dimensionally reduced data that can be visually identified by a human by principal component analysis; And obtaining the center value of the representative odor by clustering the coded odor information with reduced dimensions, and comparing the center value of the stored representative odor to obtain the degree of belonging to the center value of the representative odor of the input odor information. Judge.

Description

Method of decoding encoded information of smell information or feeling of smell transferred from video image}

1 is a view for explaining the representation and encoding of the smell information using the adjective expression factor.

2 is a diagram for explaining a method of encoding odor information according to the present invention.

3 is a diagram for explaining the application of metadata in MPEG during transmission in the method of encoding odor information according to the present invention.

4 is a graph for explaining the clustering of basic representative odors.

5 is a view for explaining a process of obtaining a center value of a representative odor by reducing the dimension and clustering of various representative odors through the PCA applied to the present invention.

6 is a diagram illustrating a transmission and decoding process of encoded data according to the present invention.

FIG. 7 is a diagram illustrating data mapped in three dimensions by the PCA algorithm method. FIG.

8 is a diagram showing the position of the unknown data by the PCA algorithm method.

<Description of main parts of drawing>

50 ... representative smell 52, 62 ... factor expression

54, 64 ... Encoding 55, 65 ... PCA

56, 66 ... clustering 58 ... Center value of representative odor

60 ... Odor Information 68 ... Representative Odor Center Values Stored

80 ... small speed determination

The present invention relates to a method of first reducing a dimension by a PCA after encoding an expression factor for a smell of an odor information or an image image, and restoring input encoding information based on a representative smell by an FCMA. B. Mapping the emotions of an odor of an image with a predetermined expression factor and encoding them, and storing some dozens of representative odors using a fuzzy clustering method to store only the central values of representative odors. Even if the received odor data comes in, the expression factors for the odor information of the odor information or image image that can be restored by obtaining the degree of belonging of the odor to be expressed by the central value stored in advance are first encoded into the PCA. By reducing the dimension and restoring the input encoding information based on the representative smell by FCMA. It is about a method.

In addition to the conventional video and audio, the Japanese Post Office decided to embark on the research and development of communication technology that delivers high-speed odors, tastes, and textures. Although communication means for appealing to sight and hearing, such as telephone and television, are common, the development of information transmission technology by human five senses of smell, taste, and touch is still an unexplored field in the world. When the five senses information communication is realized, it is expected to be applied to welfare and education fields such as telemedicine. In addition, more natural and realistic communication becomes possible.

 Five senses Information communication requires the development of technology that interprets smell, taste, and texture as information.

The sense of smell is the sensation that occurs when sensory cells in the mucous membrane of the tongue receive chemicals delivered into the nose. Many excited sensory cells are encoded into information by several or dozens of different types of sensory information receptors. These sensory information receptors can be combined in different ways, thus distinguishing so many different smells.

For example, suppose a dog called bloodhound has seven different types of olfactory receptors and can distinguish 30 different stages of stimulation from each type. Under this assumption, we must give Blothound an entire 'odor space' with 30 × 30 × 30 × 30 × 30 × 30 × 30 (= 30 ⁷ or 22 billion) distinguishable positions. It's no surprise that dogs distinguish one person from millions of people with just one smell.

In order to transmit this information through communication, development of an encoding technique and a compression technique for transmitting odor information, and a technique of accurately analyzing the information on the receiving side and reproducing so that a person feels smell and scent are essential. In other words, there is a need for a method of expressing smell in order to transmit it on a communication.

Again, according to the Japan Postal Communications Bureau, the realization of the five senses ICT is expected to be "five to ten years later." Future challenges are expected to secure communication quality over the Internet and transport technologies in wideband networks such as optical fibers. Human senses such as smell, taste, and touch should also be expressed as digital images and transmitted as in the case of visual and audio video and audio.

In general, when odor information is defined in various languages, since there are so many kinds, it is almost impossible to make a database of individual odor sensations. Accordingly, the present applicant proposes a method of mapping an emotion of an odor or an smell of an image to a predetermined expression factor, encoding the same, and restoring the input encoding information with a representative smell using FCMA. However, the immediate problem is that if the expression factor for odor is about 40-50, it is possible to express many odors by combining them, but at least 40 expression factors are determined by the way of expressing human emotions. The decision process requires a calculation in at least 40 dimensions, and the center value of the representative odor is also present in more than 40 dimensions, which requires a large memory and increases the calculation time and the calculation amount.

 The technical problem to be achieved by the present invention is to encode the expression factor for the smell of the smell information or the image image, as in the case of video and audio data by visual and auditory, and then, first, Principal Component Analysis (PCA) algorithm It is an object of the present invention to provide a method of reconstructing a large-dimensional coded data into 2-3 dimensions by using a fuzzy c-means algorithm (FCMA).

The smell information used here is defined as all the information felt in the nose. If the expression factor for odor is about 40-50, it can be expressed in combination with many odors. Even if humans have different expressions for odors, since they are clustered by common denominators of human odors, the degree of belonging to the odors is the largest. Even if two odors are mixed, it is possible to express complex odors because the degree of belonging gives a greater degree of belonging to two representative odors. According to this method, even if the expression factor is 40 or more dimensions, it can be reduced to 2 or 3 dimensions using the PCA algorithm, thus requiring only a small amount of memory for the representative smell. It can be considered that there is almost no loss of data because it can be expressed more than 98% of the original data if it is reduced to about three dimensions.

The method proposed in the present invention is expected to be applicable to MPEG-7 and MPEG-21 using meta data for the texture representation of multimedia data, and can be said to be a basic technology for standardizing odor information processing technology. have.

In order to achieve the above technical problem, the present invention provides a method of digitally decoding a smell of an odor information or an image image. The present invention performs encoding according to a representative expression of a representative odor, and clusters a center value of a representative odor by clustering. Obtaining and storing; Receiving the odor information encoded and transmitted and mapping the pattern of the data to dimensionally reduced data that can be visually identified by a human by principal component analysis; And obtaining the center value of the representative odor by clustering the coded odor information of which the dimension is reduced, and comparing the center value of the stored representative odor with the center value of the representative odor of the input odor information. It provides a digital decoding method for determining the.

Preferably, the encoding of the odor information comprises the steps of: classifying various types of language expressions that humans feel and speak as odor expression factors; Mapping a code to the classified odor expression factor; Expressing the code and concentration or degree of belonging of one or more odor expressors representing an odor in an integer or real form.

Preferably the clustering is performed using a FCMA (Fuzzy c-means algorithm), by the following equation

...수학식 12

... Equation 12

...수학식 13 a) 수학식 12에 의해 초기 소속함수 Z(0) 결정 및 클러스터 j와 패턴 i의 소속도 W_ij 을 계산하고

Equation 13 a) Determination of the initial membership function Z (0) and the membership degree W _ij of the cluster j and the pattern i by

b) calculate the cluster center Z (k + 1) by

c) ∥ Z (k) -Z (k + 1) ∥

If ∥Z (k) -Z (k + 1) ∥> ε, k = k + 1 and go back to the beginning. Otherwise, the algorithm is terminated to obtain cluster membership.

Preferably, when the actual coded smell information is transmitted, the encoded data of the smell expression factor is arranged as metadata after the image data compressed by MPEG-7 or MPEG-21.

Hereinafter, with reference to the accompanying drawings, the configuration of a preferred embodiment of the method for reducing the dimension after the encoding of the expression factor for the odor feeling of the odor information or image image of the present invention and restore the input coded odor information based on the representative odor And the operation will be described in detail.

First, prior to describing the expression factor for digitally expressing the odor information by the emotional expression of the present invention will be described the existing odor expression technology.

There have been many studies to classify and express odors, but they have suffered from difficulties due to the complexity of odors. Odors are generally more complex than color or sound, so there is no formal classification. Among other studies, Amoore published 38 basic origins and is considered one of the most common classification methods. However, Amoore's classification is not a formalized way of expressing odors properly digitally. Linda Buck, of HHMI, et al., 1999, published a report on Cell's sense of smell based on a combination of processes that recognize and process odors. Each odor receptor uses so-called receptor "alphabets" to cause specific odor responses in neurons in the brain rather than to recognize specific odors. As with language, the olfactory system uses a combination of receptors (words) to reduce the number of receptors (letters) needed to smell different odors. That is, one receptor can recognize several odors, one odor is usually recognized by several receptors, and third, different odors are also recognized by a combination of several receptors. In other words, the olfactory system uses a combination code system to smell. In other words, with different combinations, you can smell thousands of thousands of receptors. In addition, the odor enters the nose and reaches the olfactory epithelium, which is the nasal wall cells, through a hairpin-shaped rotational site. There are about 5 million olfactory neurons in the olfactory epithelium, each of which is known to express and express only one of about 1,000 olfactory receptor types. When the odor excites the neuron, its signal is transmitted through the axon to the olfactory bulb. The posterior fold is a structure located in the front of the brain, and can be said to be an information exchange organ of sense odor. And the odor signal is transmitted from the posterior brain to both the brain cortex and the limbic system, which not only conveys the actual information but also the emotional reaction.

Linda Buck's work allows us to conclude on two things.

1. A large number of odors are detected by the reactive binding code of several receptors

2. In the back of the brain, odor information and emotional reactions occur simultaneously.

The present invention uses a method of expressing various odors with less receptors as in Linda Buck's study. However, each receptor uses an adjective with an emotional meaning and is defined as an "expression factor." For example, the term "gentle" adjective corresponds to a number of odors, but if there are a lot of these adjective receptors, there are a lot of smells. Can be distinguished from. In addition, if the nature of each adjective receptor (expression factor) is divided into several, most of the smell can be classified and expressed. Table 1 below shows the basic adjective receptors of odor (odor expression factors).

Argument number Expression factor Argument number Expression factor One Thick (sexy, thick) 20 Fresh (fresh) 2 rough 21 soft 3 Masculine 22 Feminine 4 Violent 23 peaceful 5 Cruel 24 Flower smell 6 Rotten 25 Fruity 7 Fishy 26 Forest smell 8 War 27 Botanical 9 Animal 28 modern 10 Classic 29 Spicy 11 Gunpowder Smell 30 Smelly 12 fancy 31 Metal 13 Smelly 32 Stinging 14 Sperm smell 33 Park Hyang-in 15 Coffee flavor 34 Dannae 16 A man of mine 35 innocuous 17 noxious 36 Oriental 18 strong 37 Alcohol 19 Ozone 38 Disgusting

2 is a block diagram illustrating the expression and encoding of odor information using an adjective expression factor. When a certain odor information is received, an example of accepting and encoding only an expression factor of the corresponding odor information among numerous adjective expression factors is shown. As shown in Fig. 2, the odor information expression factor is digitally encoded and transmitted as transmission data. Digital encoding by odor information expression factor refers to the process of changing odor into emotional expression factor and converting it into integer type or real number type. Even if there are about 40 expression factors in total, the length of the entire code is not so large because only a few expression factors have a certain smell. The encoding converts the number and density of the corresponding expression factor into integer type or real number type, and the number of expression factors corresponding to the front part shows the length of the entire code. Also at the end is a sign indicating the odor or intensity of the image. For example, if the number of arguments is 7, the total length of the code has a length of 18 bits by adding 2x8 = 16 bits, 1 bit indicating the number of factors and 1 bit indicating the total density.

Here, when the number of arguments is n, the length m of the entire code is

m = 2 × n + 2. The length of the code does not require a separate compression algorithm, since the number of arguments may not exceed 20 at most, even if it is a complex smell or smell of smell. In addition, since only the expression factor is encoded, it may be said to include the compression effect.

As shown in Fig. 3, in the odor information encoding method, the corresponding number of factors, the number of factors, the containing speed, and the total concentration are expressed in integer form.

If it is desired to transmit the smell information by wire or wireless, it is possible to use MPEG-7 or MPEG-21 as shown in FIG. In FIG. 4, it is proposed to apply metadata in MPEG by an encoding method. For example, if an smell or image of an image is shot in a war state, the adjectives that can express it are "male", "violent", "of war", "of gunpowder," etc. The encoding is then 4 3 4 4 4 8 4 11 4. Also, if you want to express the scent of the forest from the mountain, you can express it with adjectives such as "fresh", "fresh", "soft", "forest smell", and "vegetable". However, by adjusting the concentration corresponding to each factor, the individual characteristics of the smell of acid can be distinguished. The code for this smell is 5 13 1 20 4 21 1 26 4 27 3.

It can be seen that the actual encoding length is not so large that only the corresponding expression factors are encoded, not all dozens of factors.

FCMA (Fuzzy c-means algorithm) can be used to classify and express odor information. Clustering and membership determination of odor information using FCMA can be done as follows. In order to express odor information, it is necessary to first define representative odors that exist around us. Table 2 below shows 38 basic origins determined by Amoore and socially defined odors, plus some odors not defined by Amoore.

Classification Explanation Explanation One acidic Sour incense 23 musky Musk 2 alliaceous Garlic flavoring 24 oily Oil flavor 3 almond Almond flavor 25 orange Orange flavor 4 ammoniacal Ammonia flavor 26 oxidizing Ozone flavor 5 aniseed Anise fruit flavor 27 phenolic Phenolic flavor 6 aromatic Aroma 28 putrid Corruption 7 burnt Inward fire 29 pungent Spicy flavor 8 camphoraceous Camphor 30 rose Rose scent 9 citrous Lemon-lime flavor 31 sexual Sexy incense 10 cocoa Cocoa flavor 32 spermous Sperm 11 cumin Cumin fruit flavor 33 spicy Peppermint flavor 12 edible 34 sweaty Sweat 13 ethereal Ether flavor 35 sweet Sweet flavor 14 fecal Gulin flavor 36 urinous 15 fishy Fishy scent 37 violet Violet scent 16 fruity Fruit flavor 38 woody Tree 17 green Green grass 39 coffee Coffee flavor 18 hyacinth 40 Gunpowder 19 jasmin Jasmine flavor 41 cigarette Smell of cigarette 20 lily Lily 42 ... .... 21 malty Malt flavor 43 ... .... 22 minty Peppermint 44 ... ....

In the present invention, by using the expression factors defined by Table 1, the expression factors of the respective basic smells shown in Table 2 are defined and encoded. Defining the expression arguments for basic objects is as simple as choosing the most common expression method.

However, there is a problem that the definition of the smell that humans feel may vary according to ethnicity, region, or individual taste. However, if the definition of smell commonly felt by human beings is defined without being excluded, the definition of expression differently depending on the individual may not be a big problem.

Figure 4 is a representation of each cluster after mapping some representative odors on a two-dimensional (mapping). In FIG. 4, each cluster is formed by emotions commonly felt by humans. Even though a person expresses emotions according to individual tastes without using expression factors of general feelings about the smell of group C, the other clusters are expressed in other clusters. The membership is closer to C, so it may be judged to belong to the C group. In particular, the present invention overcomes the ambiguity of human emotion by using a fuzzy clustering method (Fuzzy c-means algorithm) that exhibits robust clustering ability against ambiguity of human emotion.

In the present invention, odor information is classified and expressed using Principal Component Analysis (PCA) and FCMA (Fuzzy c-means algorithm). Clustering and membership determination of odor information using PCA and FCMA are as follows. In order to express odor information, it is necessary to first define representative odors that exist around us. However, there is a problem that the definition of the smell that humans feel may vary according to ethnicity, region, or individual taste. However, if the definition of the smell that humans feel in common is defined without being excluded, the definition of the expression that feels differently for each individual may not be a big problem.

However, since there are many kinds of odors, a lot of representative odors may occur. In particular, in the case of odors, a number of representative odors may be present because the touch and feelings to be felt in the mouth other than the five elements are to be expressed. In this case, it takes up a lot of memory because it needs to be in the decoding center of the representative odor. In particular, when FCMA is used for decoding, if there are 40 expression factors, it has to take up a lot of memory because it has to have a center value for the representative smell in 40-dimensional space. This problem may exist as a constraint in practical applications.

In order to overcome this problem, first, a method of compressing data using the PCA method, which is a method of mapping encoded data into a two-dimensional or three-dimensional space, is applied.

That is, in the present invention, the PCA method (Principal component analysis) and the FCM (Fuzzy c-means) algorithm has a correlation between the visual analysis and the quantitative analysis with the purpose of making the multi-dimensional data 2 We want to compress the data by dimming it into ~ 3 dimensions. To this end, we first obtain a three-dimensional eigenvector through the PCA method, and then implement the FCM algorithm in three dimensions using data whose dimensions are reduced by this eigenvector. Through this, even if new data comes in, it can be visually expressed, and through the center of the cluster that correlates with the visual data, the degree of belonging between each cluster of newly input data can be obtained.

5 is a diagram illustrating a clustering process for obtaining a center value of various representative scents after the dimension is reduced through the PCA. FIG. 5 encodes 54 through factor representations 52 of various representative odors 50, maps the encoded data in 2-3 dimensions through the PCA algorithm 55, and then refers to FCMA (Fuzzy c-means algorithm). The process of clustering 56 odor information by using and obtaining the center value 58 of the representative odor is shown. That is, the center value 58 of clusters can be obtained using FCMA. Here, by using the PCA algorithm, the center value for each odor can be reduced to 2-3 dimensions.

6 is a configuration diagram of a process of transmitting and decoding encoded data. In FIG. 6, when a new smell information data 60 is received, the code is transmitted through encoding 64 through the factor expression 62 of the smell, and clustered 66 to obtain the center value of the representative smell. Compared to the center value 68, since the belonging degree 70 between the clusters of the smell can be obtained, it is possible to find out which cluster the smell information belongs to. In addition, even when a plurality of odor information data is mixed, it has the advantage that it can determine which cluster is mixed through the membership analysis. This process is called odor decoding process, which also reduces the encoded data transmitted using the PCA algorithm 65 to 2-3 dimensions and compares it with the center value of 2-3 dimensions of the representative odor already stored. Express the degree of belonging of information.

Hereinafter, the PCA (Principal Component Analysis) algorithm and FCMA (Fuzzy c-means algorithm) applied to the present invention will be described in more detail.

The PCA method is an algorithm that maps a pattern to two-dimensional or three-dimensional data that can be visually identified by a human, as shown in Equation (1).

Where X is the input data received by the sensor. Where N is the number of patterns and L is the number of sensors. This N × L data is now changed to the N × D data form as shown in Equation (2).

Where Y is mapping data and D is the number of dimensions desired. If k = 1, 2, ..., n, Equation 3 is used to calculate the average for each sensor.

를 빼어감에 따라 계산되어 진다.Where n is the number of sensors and m is the number of patterns. Using the average data, Equation 4 is used to obtain a covariance matrix for the input. The covariance matrix C is the mean vector in the input pattern X so that each element, C _ij, is _centered over all classes.

Calculated by subtracting

Next, eigenvalues λ _k and eigenvectors u _k for k = 1, 2, ..., n are obtained, and then defined as in Equation 5.

Select the largest eigenvalues λ ₁ , λ _2, and λ ₃ from the calculated λ _k (if 3-D mapping), and use the eigenvectors u ₁ , u _2, and u ₃ as the principal components for multidimensional data. Mapping in three dimensions.

In general, mapping in three dimensions allows some mapping without significant loss of data.

This PCA method allows for clustering analysis visually. However, when new data is acquired, a separate algorithm is required to quantitatively determine which cluster is close to or belongs to which cluster.

To this end, in the present invention, three-dimensional data obtained based on the three-dimensional eigenvectors obtained using the PCA algorithm are obtained from the center of the cluster using the FCM algorithm, and the data inputted using the obtained center of the cluster. Apply the method of obtaining membership of.

Basically, the FCM algorithm is unsupervise learning, but if you predefine standard data on odor, it can be determined as the same cluster for the same data. It is possible.

Quantifying the expression using a separate k-means algorithm or FCM algorithm has a different correlation between the visually analyzed data and the data obtained by the FCM algorithm because the dimensions of the PCA method are different from those of the FCM algorithm. Have. The use of the FCM using the results of the PCA enables to obtain the degree of belonging between each cluster of newly input data through the center of the cluster that correlates with the visual data represented by the PCA.

When quantitative analysis of the FCM algorithm to display the elements in the Y _i haeo the data mapped to the PCA process to an input vector by a vector y

Where n is the number of input patterns and represents a matrix reduced in three dimensions by PCA. If it is displayed again as an input pattern, it is as shown in Equation 7-1.

Equation 7-1

The center of the cluster for the input pattern set is represented by Equation 7-2 below.

[Equation 7-2]

Where c is the number of cluster centers, _i of i is the number of input patterns, and _j of j is the number of clusters. Therefore, the membership degree W _ij of the center of the i-th pattern and the j-th cluster is expressed as follows.

Each of the above patterns may have different membership and clusters. As shown in Equation 9 and Equation 10, the sum of the degree of belonging to one pattern is normalized to 1, and each pattern belonging degree finds a cluster pattern in which the objective function J _m is minimum.

D _ij in the objective function J _m is the Euclidean distance between the input pattern and the center of the cluster. This measures the similarity for each pattern. Where m is a membership weighting exponent and m 〓 2 is usually chosen because there is no theoretical and justified rule for choosing this value.

The solution of the objective function of the FCM algorithm proposed by Bezdek is as follows.

FCMA is an iterative algorithm that approximates the minimum value of the objective function of Equation 13 without being solved analytically by Equations 12 and 13. Therefore, the learning values of equations 12 and 13 converge to optimal values.

Since this algorithm is an unsupervised pattern recognition technique, a corresponding learning procedure is required. This learning procedure brings the degree of belonging to the cluster center of the input pattern and indicates the distance of each input pattern to the cluster center.

The process of FCM algorithm for learning is summarized as follows.

1. Determination of initial membership function Z (0) and calculation of membership degree W _ij of cluster j and pattern i through Equation 12

2. Calculation of cluster center Z (k + 1)

3. Calculation of Z (k) -Z (k + 1) ∥

If ∥Z (k) -Z (k + 1) ∥ε = k = k + 1, go back to the beginning, otherwise terminate the algorithm.

As described above, by clustering the results of the dimension reduction through the PCA process through the FCM algorithm, new sense of belonging, visual effects, and quantitative results of the existing data can be obtained simultaneously.

Through this process, the center of the cluster with respect to the standard patterns obtained in advance and the cluster membership of each standard pattern can be obtained. The cluster centers obtained here are stored in the computer in advance and used to calculate the degree of belonging to an unknown sample when an unknown sample is input.

Where S is an unknown sample vector and j = 1, 2,... , c and the degree of belonging between the center of the standard pattern and the sample pattern can be obtained by the following equation (15).

In Equation 15, V _j represents a degree of belonging of the sample vector to the j th cluster.

Table 3 below shows the odor data obtained by three questionnaires, and the data for encoding seven basic odor expression factors by three persons determined as samples and then encoding them.

One 2 3 4 5 6 7 8 9 100 11 12 13 14 15 16 17 181 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 Tenderness 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 40 10 0 0 0 0 0 0 0 5 0 100 0 300 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 42 0 12 0 0 0 0 0 0 0 6 0 100 0 32 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 45 0 10 0 0 0 0 0 0 0 5 0 100 0 30 0 0 0 0 0 0 Garlic flavor 0 20 35 0 0 0 0 0 0 0 0 0 20 0 0 0 0 60 0 0 0 0 0 0 0 0 23 0 60 0 0 39 0 0 0 0 0 0 0 22 40 0 0 0 0 0 0 0 0 0 15 0 0 0 0 62 0 0 0 0 0 0 0 0 20 0 62 0 0 42 0 0 0 0 0 0 0 27 27 0 0 0 0 0 0 0 0 0 18 0 0 0 0 59 0 0 0 0 0 0 0 0 22 0 65 0 0 43 0 0 0 0 0 0 Ammonia flavor 0 0 0 0 100 90 0 0 0 0 0 0 0 0 0 0 80 50 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 0 0 0 0 100 92 0 0 0 0 0 0 0 0 0 0 85 55 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 0 0 0 0 100 95 0 0 0 0 0 0 0 0 0 0 75 45 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 Aroma incense 0 0 0 0 0 0 0 0 0 0 0 80 0 0 0 0 0 0 0 10 60 80 90 100 10 25 90 30 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 85 0 0 0 0 0 0 0 13 65 82 95 100 8 20 92 20 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 90 0 0 0 0 0 0 0 11 60 85 100 98 9 22 89 25 0 0 0 0 0 0 0 0 0 0 Bullet 0 50 0 0 0 0 0 0 0 0 0 0 0 0 0 100 70 0 0 0 0 0 0 0 0 0 0 0 20 0 0 0 0 0 0 0 0 0 0 55 0 0 0 0 0 0 0 0 0 0 0 0 0 100 72 0 0 0 0 0 0 0 0 0 0 0 15 0 0 0 0 0 0 0 0 0 0 60 0 0 0 0 0 0 0 0 0 0 0 0 0 100 79 0 0 0 0 0 0 0 0 0 0 0 22 0 0 0 0 0 0 0 0 0 Lemon Lime 5 0 0 0 0 0 0 0 0 0 0 20 0 0 0 0 0 0 0 40 0 0 20 15 100 20 70 60 0 35 0 0 0 30 0 0 0 0 5 0 0 0 0 0 0 0 0 0 0 22 0 0 0 0 0 0 0 39 0 0 19 15 100 18 70 62 0 37 0 0 0 29 0 0 0 0 7 0 0 0 0 0 0 0 0 0 0 18 0 0 0 0 0 0 0 38 0 0 21 16 100 17 73 65 0 30 0 0 0 33 0 0 0 0 Smell of cigarette 0 80 70 30 0 20 0 0 0 0 10 0 0 0 0 70 90 80 0 0 0 0 0 0 0 0 0 10 20 0 0 0 0 0 0 0 0 30 0 82 72 35 0 18 0 0 0 0 11 0 0 0 0 72 92 82 0 0 0 0 0 0 0 0 0 10 18 0 0 0 0 0 0 0 0 31 0 86 78 40 0 22 0 0 0 0 12 0 0 0 0 80 89 88 0 0 0 0 0 0 0 0 0 12 22 0 0 0 0 0 0 0 0 30

The seven odors in Table 3 are graphically represented in FIG. 7 shows data mapped in three dimensions by the PCA method. In addition, as shown in Fig. 8 and Table 4, after having a person in charge of ammonia and lemon scent, after determining the expression factor, it is encoded, it is encoded by the cluster center predetermined by the FCM Was determined.

Smell of the Unknown 1 Unknown Smell 2 Affiliation 0.0321 0.0023 Belonging to garlic flavor 0.0326 0.0024 Position of ammonia flavor 0.8251 0.0016 Affiliation of Aroma 0.0216 0.9857 Affiliation of Tan inward 0.0313 0.0017 Lemon-Lime Flavor 0.0272 0.0045 Position of tobacco smell 0.0302 0.0019

Through this experiment, it was found that even people with different emotional judgments can decode odor data by the FCM algorithm without any significant difference from predetermined general emotional data. As shown in Table 4, in case of unknown odor 1, it has about 83% of ammonia belonging, and other data can be judged perfectly as ammonia because it is not more than 3%. In addition, by mapping all the data in three dimensions before using the FCM algorithm, the computation time is reduced, and the dimension of the representative odor is reduced to require less memory or a small amount of transmission.

As described above, after encoding the expression factor for the odor feeling of the odor information or the video image of the present invention, the method first reduces the dimension by the PCA and restores the input encoding information based on the representative smell by the FCMA. It is possible to reduce the capacity of the memory for storing the representative smell.

As described above, in the case of video and audio data according to the visual and auditory aspects of the present invention, after encoding the expression factors for the smell of smell information or the smell of the video image, the dimension is first reduced by PCA and then The method of restoring the input encoding information based on the representative smell can restore the smell information without occupying much memory. However, the method of expressing the expression factors presented in the present invention is not limited thereto, and it will be apparent to those skilled in the art that various modifications may be made, and thus the description of each modification will be omitted. In the present invention, the part for sensing the odor information is not described in detail, but when the odor information processing technology of the present invention is connected to a sensor for sensing the odor information, the odor information can be processed digitally, and the odor information can be remotely controlled. There is an effect that can be judged.

Claims (4)

In the method of digitally decoding the smell of the smell information or the image image, Performing encoding according to the factor expression of the representative odor and obtaining and storing the center value of the representative odor by clustering; Receiving the odor information encoded and transmitted and mapping the pattern of the data to dimensionally reduced data that can be visually identified by a human by principal component analysis; And Obtain the center value of the representative odor by clustering the coded odor information with reduced dimensions, and compare the center value of the stored representative odor with the center value of the representative odor. Digital decoding method for judgment. The method of claim 1, wherein the encoding of the smell information is Classifying various forms of verbal expression as odor expression factors by which humans feel and speak odors; Mapping a code to the classified odor expression factor; A method of digitally decoding odor information comprising the step of expressing the code and the concentration or the degree of belonging of one or more odor expression factors expressing one odor in an integer or a real number form. The method of claim 1 or 2, wherein the clustering is performed by using a fuzzy c-means algorithm (FCMA),

... Equation 12

... Equation 13 a) Determining the initial membership function Z (0) and calculating the membership degree W _ij of the cluster j and the pattern i by using Equation 12 b) calculate the cluster center Z (k + 1) by c) ∥ Z (k) -Z (k + 1) ∥ If ∥Z (k) -Z (k + 1) ∥> ε k = k + 1 and go back to the beginning, otherwise, the algorithm is terminated to obtain cluster membership. The method according to claim 1 or 2, wherein when the actual coded smell information is transmitted, encoded data of the smell expression factor is arranged as metadata after the image data compressed by MPEG-7 or MPEG-21. Digital information decoding method for odor information.

Images