JP2573438B2 - Pattern identification method and apparatus - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for pattern identification.

[0002]

2. Description of the Related Art As one genre of pattern identification technology, template matching is widely known, but there have been the following problems with this template matching.

That is, when performing multi-valued template matching using the density distribution of an image to be identified as it is, it is easily affected by illumination at the time of image reading, and the similarity or dissimilarity differs when the illumination state changes. Also, the discrimination ability is reduced due to a change in contrast of the discrimination target image itself.

[0004]

SUMMARY OF THE INVENTION The present invention has been made in view of the above-described circumstances, and improves the conventional template matching to be influenced by illumination conditions and changes in contrast of an image to be identified. It is an object of the present invention to provide a new and difficult pattern identification method and an apparatus for implementing the method.

[0005]

The pattern identification method according to the first aspect of the present invention is a conventionally known template matching type identification method.
"A comparison pattern is created by replacing the density of each pixel of the identification pattern cut out to the same size as the template with the density of the template element according to the density order, and the similarity or dissimilarity between the comparison pattern and the template. Calculate the degree. "

[0006] The pattern identification method according to the second aspect of the present invention.
Is an embodiment of the method of claim 1,
"Image input process, cropping area setting process, and cropping
Process, density order table creation process, density conversion process,
Matching step and determination step ".

The method will be described with reference to FIG. 1 which is a flowchart showing the method. The “image input step” is a step of reading the image to be identified and storing the density of each pixel as a digital signal.

[0008] The "cutout area setting step" is a step of setting a cutout area and a cutout start position as areas for performing pattern identification on the image information stored in the image input step.

[0009] The "cutting step" is performed from the cut area.
This is a step of cutting out an identification pattern of the same size as the template.

The “density order table creation step”
The density of each pixel of the identified pattern is rearranged in the order of density
Then, a density order table is created.

The “density conversion step” is performed in the order of the produced density.
Table and template density order table prepared in advance
, The density of each pixel of the identification pattern sequentially corresponds to
Replaced with the values in the ranking template density order table,
This is a step of creating a comparison pattern.

In the flowchart of FIG . 1, this corresponds to the fourth step.
The “density conversion” is the same as the “density order table creation process” described above.
"Density conversion step" is shown together.

The “matching step” includes a comparison pattern,
This is a step of calculating the similarity or the dissimilarity with the template.

[0014] The above-described "from the cutting step to the matching step" is repeated over the cutting area by changing the cutting position of the identification pattern.

The "determining step" is a step of determining whether or not a pattern identical or similar to the template exists in the cut-out area based on the calculation result of the matching step.

In the method according to the second aspect, the "concentration"
In the degree order table creation process, the identification pattern
The density and address of each pixel of the pattern in the order of density.
And the density order image address table together with the density order table.
Bull "can be created. In this case,
Degree conversion process is "Density of each pixel of the created density order table
According to a predetermined density numbering method.
Concentration number table creation process to convert to
Each table element of the density number table
To indicate the order of the elements in the template density order table
Read the density of the element as a dex ,
Write to the address specified by the above table element
Then, a comparison pattern creation process to create a comparison pattern "
(Claim 3).

According to a fourth aspect of the present invention, in the pattern identification device,
Apparatus for implementing the method of claim 2, wherein "reading means"
, Storage means, cutout means, density order table creation
Means, density conversion means, calculation means, control identification means
And ".

The "reading means" is means for reading an image to be identified. The reading means is a television camera or a one-dimensional sensor, a two-dimensional sensor or the like, and converts the identification object image, means for converting the density of each pixel into an electrical signal, the signal converted into the "digital signal" A / It can be configured with a D converter. For example, the A / D converter, by using those 8 bits, can be multi-valued in a concentration of 256 (= 2 ⁸⁾ stages of each pixel. "Storage means", the concentration of each pixel of the reading obtained identification object image, a means for and storing the digital signal.

The "cutout means" sets a cutout area as a pattern identification area and a cutout start position with respect to the image information stored in the storage means, and sets the same size as the template from the set cutout area. the identification pattern is a means for cutting.

The "density order table creating means" rearranges the density of each pixel of the cut-out identification pattern in the order of density.
Means for creating a density order table.

The "density converting means" includes a created density order table and a prepared template density order table.
, The density of each pixel of the identification pattern is sequentially corresponded.
Replaced with the values in the template density order table
Means for creating a comparison pattern.

The "calculating means" is a matching device for calculating the similarity or dissimilarity between the comparison pattern and the template.
It is a means for performing the process. Perform similarity or dissimilarity
The “operation expression” to calculate
Various formulas known for the switching can be appropriately used .

The "control identification means" is a match of the arithmetic means.
Based on the calculation result of the
"Same or similar pattern as template" exists
This is a means for determining whether or not to perform .

[0024] In the pattern identification apparatus according to the fourth aspect,
In the density order table creation means,
The density and address of each pixel of another pattern are re-ordered in the order of density.
Arrange, density order table and density order image address table
And create it. " In this case, the density conversion
Means is that "each pixel density of the created density order table
Convert to density number array according to the specified density numbering method
Means for creating density number table "and"
Each table element of the number table is
Index indicating the order of elements in the plate concentration order table
Read the density of the element as the
Write to the address specified by the
Comparison pattern creating means for creating a comparison pattern "
(Claim 5). like this
Then, the pattern identification method according to claim 3 is implemented.
can do.

The above storage means, cutting out means, density
Order table creating means, density converting means, calculating means and
Computer and computer identification means.
These can be made up of data and additional memory .

[0026]

FIG. 2 is a functional block diagram of the pattern identification device according to the fifth aspect . As shown, the identification object image is read by the image input unit 21 as a video signal, by the A / D converter 22, is converted into a density is de digital signal for each pixel, is stored in the image memory 23. The image input unit 21 is a television camera or the like as described above.
Together with the / D converter 22, the "reading means" is constituted.
The image memory 23 constitutes the "storage means" .

The image cut-out unit 24 , which constitutes “cut-out means” , sets the “cut- out area ” as an area for performing pattern identification on the image information stored in the image memory 23.
And an extraction start position, and an identification pattern having the same size as the template is extracted from the set extraction area. In other words, the image cut-out section 24, "cut out territory
Area setting step "and" cutout step " .

FIG. 3 is an explanatory diagram showing the setting of the cut-out area and the cut-out process. The area 30 indicated by the broken line is the “identification target” stored in the image memory 23 as storage means.
This is the data area of the “entire image” . During this data area 30, a region 31 shown by the solid line is "cut-out region". That is, the image information in the cutout area 31 is a target for matching.

The hatched rectangular area 32 drawn in the cutout area 31 is the "identification pattern" to be cut out. The identification pattern 32 has the same size as the template, and is cut out one after another in a predetermined order from the cutout start position to the cutout end position, and each of the cutout identification patterns is a target for matching with the template pattern. Becomes

Returning to FIG. 2, the cut-out identification pattern is created in a "density order table" by a density order table creation unit 25 constituting "density order table creation means" . That is, in the cut-out identification pattern, address and density information of each pixel (digital information multivalued)
Are sorted in ascending or descending order of density.

[0031] sorted concentration information is re-arranged according to the order in which they are sorted, "concentration order tables" is created.

In the method according to the second aspect, the density order table
Only the above density order table is created by the creation process
However, in the method according to claim 3, the concentration order tape
While the file is created, the "address" is also sorted
Are arranged in the order in which
Bull "is created. These two tables are
The matrix corresponds to the density and
The dress is the same as “each pixel address and
And the density of the pixel.

A state in which a density order table and a density order image address table are created from the identification pattern is schematically shown in the upper part of FIG. Of course, the "image address", the cut identification pattern, the address information of each pixel.

Referring to FIG. 4, a "density number table" is created from the density order table created as described above. The density number table is created by converting each pixel density in the density order table into an “array of density numbers” according to a predetermined density number assignment method. The details of this step will be described later in the description of the embodiments.

Each "table element" of the density number table thus created is a "density number". This concentration numbers, on the one hand, while being compared with the template concentration Junte over Bed <br/> Le (are prepared with pre-template), on the other hand, used in "the specified address position of the comparison pattern". That is, each density number is stored in the density order image address table at the address corresponding to the address of the density number in the density order image address table.
Address is selected as the address of the comparison pattern.

On the other hand, the density number, the template concentration sequence table, to specify the concentration order of template concentration Junte Buru in "the same address as the address of this concentration Number", the concentration of template elements designated by the designation (In the arrangement order in which the elements of the template are arranged in the order of density, the density order specified above is used as the array order.
Template element density).

In other words, the created density number table
Each table element of "Template density order table
This is used as an index indicating the order of elements.
The density of the element corresponding to the index is the template density
Is read from the sequential table, and the
It is written to the specified address.

At the address determined as described above,
When writing the selected density is performed “for all table elements of the density number table”, a comparison pattern is obtained. I made this comparison pattern into a table
This is the comparison table of FIG.

This "comparison pattern" represents the density array of the identification pattern using the density of the template element. Since the density array of the identification pattern is converted, the above process is performed by the density conversion. It is called a process.

FIG. 5 is a flowchart showing a process from the creation of the density order table to the creation of the comparison pattern.
"Density value reading of template by density number" in the figure
Means that “the density of the template element having the order of the density specified by the density number (the above index) is selected” described above , and “writing the value to the position indicated by the image address table”. "
This means "writing the selected density to an address determined according to the density number". The determination of “table end?” Is for determining whether or not the writing of the comparison pattern has been completed.

Returning to FIG. 2, the process from the above-described density number table creation step to the creation of the comparison pattern is performed in the density conversion unit 26. That is, the density converter 26 constitutes a density number table creating means and a density converting means.
Then, in order to create a comparison pattern, template concentration Junde chromatography data (and the template concentration order table, the data of the density corresponding to the respective density ranks) to input a 29-A.

Next, a matching step for calculating the similarity or dissimilarity between the comparison pattern created as described above and the template is performed. This matching process is performed by the calculation unit 27 constituting the calculation means. As the “calculation of similarity or dissimilarity” performed here, various calculation methods conventionally known as described above can be appropriately used.

FIG. 6 schematically shows a comparison pattern and a template. T the information content of the template
(U, v), the information content of the comparison pattern is f (u, v)
(U, v the address of each information) When, for example, arithmetic expression: Σ | f-t | (the sum, taken over the entire address u, v) and Σ (f-t) ² the "dissimilarity It can be used as a ". The greater the calculation results of these arithmetic expressions, a high degree of dissimilarity between the comparison patterns and the template Karadea
You. In addition, an arithmetic expression: Σ (f · t) (the sum is the address u,
v) can be used as "similarity"
You.

Returning again to FIG. 2, the arithmetic section 27 stores information on the comparison pattern created in the density conversion step,
The template data 29-B prepared in advance is taken in, and the dissimilarity is calculated by the above calculation.

The above process is repeated as necessary by sequentially changing the identification patterns to be cut out. In FIG. 1, the determination step of “region end?” Is a step of determining whether or not the above-described density conversion and the matching process have been performed by extracting all the identification patterns to be extracted over the entire extraction region.

Subsequently, a "determining step" is performed. This step is executed by the control identification unit 28 constituting the control identification means in the example of FIG. The determination is made by checking whether or not the degree of dissimilarity calculated for each identification pattern in the matching step exceeds a predetermined threshold, and if it exceeds the threshold, there is no pattern similar to the template in the cutout area. And in the opposite case, it is determined that a similar pattern exists.

[0047]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, as specific examples, a case where a cut-out identification pattern and a template are similar and a case where they are not similar will be described.

The embodiment shown in FIG. 7 is an example in which the cut-out identification pattern and the template are similar. That is, FIG. 7E shows a template. The template is composed of 4 × 4 16-pixel multivalued digital data. This template data
The “rearranged in the order of density” is shown in FIG.
It is a template density order table.

[0049] corresponding to each element of the template concentration order table, shown in those are arranged numbered from 0 in order until 15 FIG (F-2) "template concentration order tape
Is a bull ", the above-mentioned template concentration Junde chromatography data (Fig. 2
29-A) is constituted by FIGS. 7 (F-1) and 7 (F-2).

Now, in FIG. 7, (A-1) and (A-
2) is a cut-out identification pattern (consisting of 4 × 4 16 pixels like the template),
The address is shown. FIG. 7 (B) shows that the identification patterns and the addresses are rearranged in the order of decreasing density.
1B is a density order table and FIG. 2B is a density order address table. FIG. 7C shows a density number table in which each density value in the density order table is replaced with a density number according to a predetermined density number assignment method.

The density number is the number of pixels constituting the template:
According to No. 16, 0 to 15 are prepared with the maximum density being 0. Now, assuming that a density number: i (i = 0 to 15) is given to the density at the n-th (n = 1 to 16) address in the density order address table, the n-th density M consecutively from the sequential address table
When the same density continues for the addresses, these addresses: n, n + 1, n + 2,. . , N + m−1, the density number: i is given, and the address: n +
For the density of m, the density number: i + m is given. This situation is the same as when there are two first places in a sports race, the first place is two and the second place is missing, and the next place is the third place.

Of course, the method of assigning the density number is not limited to this example, and another method may be used.

Each element of the density number table (C) created in this manner is used for specifying an address of a comparison pattern (FIG. 7D) and selecting density information to be written at the specified address. . Hereinafter, creation of the comparison table based on the density number table will be specifically described.

Attention is paid to the element at the upper left corner of the density number table shown in FIG. This element has a density number “0”. When the same address as this element (upper left corner of the table) is applied to the density order address table (B-2), the same element on the density order address table has address: 2 as its content. This address: 2
"Address in comparison pattern" is determined. That is,
The element at the upper left corner of the density number table specifies the address: 2 in the comparison pattern, that is, the second cell from the left at the top of the table in FIG. 7D.

Next, when the density number “0” of the element at the upper left corner of the density number table is used as an index and searched in the template density order table, a density order {0} having the same value as the density number “0” is obtained. It can be seen that the template density value: 16 corresponds (see template density order table (F-1)).

Therefore, the above-mentioned density value: 16 is written at address: 2 in the comparison pattern shown in FIG. 7D. For each element of the concentration number table,
When a comparison pattern is written by repeating the above process, a comparison pattern as shown in FIG. 7D is created.

In the conventional template matching, the cut-out identification pattern (A-1) and the template (E) are compared to calculate a similarity or a dissimilarity. The similarity / dissimilarity between the pattern (D) and the template (E) is a problem. It should be noted here that in this embodiment, the identification pattern (A-1) and the template (E) are similar but not identical. However, when the identification pattern is converted into the comparison pattern by the density conversion, the comparison pattern and the template are “the same pattern” as shown in FIG.

An embodiment in which the identification pattern and the template are dissimilar is shown in FIG. 8 following FIG. In this case, the comparison pattern (FIG. 8 (D)) and the template (E) obtained by the density conversion are completely different patterns.

From this, the following can be said. That is, in the density numbering method in this example, when the identification pattern is similar to the template pattern, there is a property that a comparison pattern more similar to the template pattern is created by density conversion. This improves the efficiency and accuracy of matching.

[0060]

As described above, according to the present invention, it is possible to provide a novel pattern identification method and apparatus in the template matching method. The present invention, as described above,
Since the cut-out identification pattern is not directly compared with the template, the comparison pattern obtained by converting the identification pattern using the density of the template is compared with the template,
Even if the illumination condition of the identification target image changes, or the absolute contrast of the identification target image itself changes, proper pattern identification is possible as long as the relative contrast of the identification target image can be correctly signaled. .

The "density conversion" used in the present invention
As such, it can be applied not only to matching but also to neural network operations and the like.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining a process of the present invention.

FIG. 2 is a functional block diagram showing an example of the device according to claim 3;

FIG. 3 is a diagram for explaining extraction of an identification pattern.

FIG. 4 is a diagram for explaining density conversion.

FIG. 5 is a flowchart for explaining density conversion.

FIG. 6 is a diagram for explaining a non-similarity calculation between a template and a comparison pattern.

FIG. 7 is a diagram for explaining one embodiment.

FIG. 8 is a diagram for explaining another embodiment.

Claims (5)

(57) [Claims] 1. A pattern identification method of a template matching method, wherein the density of each pixel of an identification pattern cut out to the same size as a template is replaced with the density of a template element in accordance with the density order. And calculating the similarity or dissimilarity between the comparison pattern and the template. 2. An image input step of reading an image to be identified and storing the density of each pixel as a digital signal, and, based on the stored image information, a cutout area as a pattern identification area and a cutout start position. A cutout area setting step to be set; a cutout step of cutting out an identification pattern having the same size as the template from the cutout area; and a density of each pixel of the cutout identification pattern,
Density table to create a density order table
Template creation process , created density order table, and prepared template
The density of each pixel of the identification pattern is
Template density order table of the order corresponding to the degree
A density conversion step of creating a comparison pattern by replacing the value of the comparison pattern; a matching step of calculating the similarity or dissimilarity between the comparison pattern and the template; and a template in the cutout area based on the calculation result of the matching step. A determination step of determining whether or not a pattern identical or similar to the pattern exists. 3. A pattern identification method according to claim 2, wherein
In the density order table creation process, the identification
Relocation turn, the concentration and the address of each pixel in the concentration order
And a density order table, a density order image address table,
Is created, and the density conversion step converts each pixel density of the created density order table into a predetermined density.
Density number to be converted to density number array according to numbering method
No. table creation step and each table element of the created concentration number table
Change the order of the elements in the prepared template density order table.
Read the density of the element as an index to indicate
Density to address specified by the above table element
Write a comparison pattern to create a comparison pattern
And a process for identifying a pattern. 4. An apparatus for carrying out the method of claim 2, wherein: reading means for reading the image to be identified; storage means for storing the density of each pixel of the read image to be identified as a digital signal; For the image information stored in the storage means, a cutout area as a region for performing pattern identification and a cutout start position are set, and a cutout means for cutting out an identification pattern having the same size as the template from the set cutout area, A density order table creating means for creating a density order table by rearranging the density of each pixel of the cut out identification pattern in the order of density, a created density order table, and a template density order table prepared in advance. , The density of each pixel of the identification pattern is sequentially replaced with the value of the template density order table of the corresponding order, and the comparison pattern Density converting means to be created; calculating means for performing a matching step for calculating the similarity or dissimilarity between the comparison pattern and the template; Alternatively, a pattern identification device having control identification means for determining whether or not a similar pattern exists. 5. The pattern identification device according to claim 4, wherein the density order table creating means rearranges the density and address of each pixel of the cut-out identification pattern in the order of density,
A density order table and a density order image address table; and a density conversion means for converting each pixel density of the created density order table into an array of density numbers according to a predetermined density number assignment method. Using each table element of the created density number table as an index indicating the order of the elements in the prepared template density order table, read the density of the element, and read the density to the address specified by the table element. 4. A pattern identification device for performing a pattern identification method according to claim 3, further comprising comparison pattern creation means for writing and creating a comparison pattern.