JP3104355B2 - Feature extraction device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for recognizing images such as characters.

[0001]

2. Description of the Related Art In order to perform processing for recognizing characters and symbols, differences in the shapes of the characters and symbols are well represented.
It is necessary to extract features that are not easily affected by deformation and noise from the target image.

[0002] Characters and symbols can be broken down into small areas to form vertical, horizontal, or diagonal lines. The direction of the line in each area is a characteristic of the character or symbol.
It is effective for its recognition. To avoid being affected by the line thickness, it is effective to extract features only from the outline. For example, "Identification of handwritten kanji by contour direction contribution characteristics" (IEICE Transactions (D) Sho5
8-Theory 488 [D-143] Hagita, Naito, Masuda) Scans the image in multiple directions from the contour pixels, checks the number of consecutive black pixels, and characterizes the ratio of each connected number Or JP-A-63-293691 or JP-A-6-29391.
As disclosed in Japanese Patent Application Laid-Open No. 3-219089, there is known a method in which the state of surrounding pixels is examined in the outline of a character, the direction is determined, and the feature is determined.

[0003]

However, in the method described in the above-mentioned paper, the numerical value of the feature is represented by a decimal number, and the numerical value in each direction must be held as a feature. The equipment becomes complicated,
It takes time to extract and compare features.

Further, in the method disclosed in the above-mentioned publication, only the state of peripheral pixels is reflected, so that the direction of a line constituting an actual character or symbol is not represented, or a fine outline portion due to noise or the like is not expressed. They may be adversely affected by unevenness. In particular, many pixels in the contour portion have a density close to the value at the time of binarization, so that the pixel is determined to be a white pixel or a black pixel due to slight noise, so that a lot of fluctuation such as unevenness occurs. I do. Further, it is difficult to parallelize the process of tracking the contour, and it takes time to extract features.

[0005] The present invention solves the above problems,
Provide a means for recognizing characters and symbols with a high literacy rate and in a short time by a simple and parallel processing method for quickly obtaining features that effectively represent the positions and directions of lines constituting characters and symbols. This is an object of the present invention.

[0006]

According to a first aspect of the present invention, continuity in a plurality of directions is checked for each pixel of an input image, and a direction continuity value representing the continuity is given to each direction. Direction continuation value providing means, direction attribute determining means for determining which direction attribute the pixel of interest has based on the direction continuity value of each pixel obtained by the direction continuous value providing means, and determination by the direction attribute determining means Direction attribute counting means for counting pixels having the determined direction attribute for each direction attribute.

According to a second aspect of the present invention, a direction continuity value assigning means for examining continuity in a plurality of directions for each pixel of an input image and assigning a direction continuity value indicating the continuity to each direction, A direction continuous value diffusing unit for performing a process of diffusing the direction continuous value of each pixel obtained by the direction continuous value providing unit in a space corresponding to the input image to change the direction continuous value; A direction attribute determining means for determining which direction attribute the pixel of interest has based on the direction continuous value of each pixel changed by the above, and a pixel having the direction attribute determined by the direction attribute determining means for each direction attribute And a direction attribute counting means for counting.

[0008] A third invention is a direction continuity value assigning means for examining continuity in a plurality of directions for each pixel of an input image, and assigning a direction continuity value indicating the continuity to each direction, Direction attribute determining means for determining which direction attribute the pixel of interest has based on the direction continuous value of each pixel obtained by the direction continuous value providing means, and an area for dividing a space corresponding to the input image into a plurality of areas Dividing means, and direction attribute counting means for counting, for each direction attribute, pixels having the direction attribute determined by the direction attribute determining means in the area divided by the area dividing means. I do.

A fourth invention is directed to each pixel of a force image,
Check the continuity in a plurality of directions, a direction continuity value providing means for providing a direction continuity value representing the continuity to each direction, and a direction continuity value of each pixel obtained by the direction continuity value providing means, A direction continuous value diffusion unit that performs a process of diffusing in a space corresponding to the input image and changes the direction continuous value, and in which direction the target pixel is determined based on the direction continuous value of each pixel changed by the direction continuous value diffusion unit. Direction attribute determining means for determining whether or not the area has an attribute; area dividing means for dividing a space corresponding to the input image into a plurality of areas; and in the area divided by the area dividing means, determination is made by the direction attribute determining means. Direction attribute counting means for counting pixels having the determined direction attribute for each direction attribute.

[0010]

According to the first aspect of the invention, each pixel of the input image is checked for continuity in a plurality of directions by the direction continuity value providing means, and a direction continuity value indicating the continuity is provided to each direction. Is done. Next, the direction attribute determining unit determines which direction attribute the target pixel has based on the direction continuous value of each pixel obtained by the direction continuous value providing unit. Finally, in the direction attribute counting unit, pixels having the direction attribute determined by the direction attribute determining unit are counted for each direction attribute. Therefore, since the direction continuation value of each pixel of the input image is determined, it is possible to obtain a feature including the global position and direction of the line constituting the input image.

Further, the second invention is different from the first invention in that
The directional continuity value providing unit performs a process of diffusing the directional continuity value of each pixel obtained in the directional continuity value providing unit in a space corresponding to the input image, and includes directional continuity value diffusion unit for changing the directional continuation value. Can be diffused by the direction continuous value diffusion means.

Further, the third and fourth aspects of the present invention are different from the first and second aspects in that an area dividing means for dividing a space corresponding to the input image into a plurality of areas is provided, and the area divided by the area dividing means is provided. The pixels having each direction attribute are counted in the above, so that the feature can be extracted more accurately.

[0013]

Embodiments of the present invention will be described below with reference to the drawings.

In FIG. 1, reference numeral 101 denotes an image input device which optically reads a document or the like, performs photoelectric conversion, and uses the characters and symbols as image data. Reference numeral 102 denotes a preprocessing device that performs processing such as binarization and noise component removal on an input image, sequentially detects a unit area corresponding to one character from the image, and outputs image data in the area. It is. 10
Reference numeral 3 denotes a feature extraction device which forms a main part of the present invention. 104
Is a dictionary pattern storage device that stores features extracted from a standard character image in advance using the preprocessing device 102 and the feature extraction device 103 in correspondence with character codes. Reference numeral 105 denotes a feature comparison device that compares the feature pattern of the character image output from the feature extraction device 103 with the features of a plurality of standard character images in the dictionary pattern storage device 104, and outputs a character code according to the result. .

FIG. 2 shows the internal configuration of the feature extracting device 103 which is a main part of the present invention. Reference numeral 201 denotes a direction continuity value assigning unit that examines the continuity of each pixel of an image in a plurality of directions with other pixels that are continuous with the direction, and assigns a direction continuity value indicating the continuity to each direction. is there. Reference numeral 202 denotes a direction continuous diffusion unit that performs a process of diffusing the direction continuous value for each direction given to each pixel, and corrects the direction continuous value of each pixel. A direction attribute determination unit 203 determines whether each pixel has the direction attribute based on the direction continuous value processed by the direction continuous diffusion unit 202. An area dividing unit 204 divides a unit area occupied by one character into a number of elementary areas in a prescribed shape. A direction attribute counting unit 205 counts the number of pixels having each direction attribute determined in each element region. Reference numeral 206 denotes a direction attribute collection unit that collects the direction attribute count values of the image of each elementary region in the unit region occupied by one character and extracts an image feature pattern.

Now, the operation of the feature extracting device 103 will be described in detail. It is assumed that the image input to the feature extraction device is binarized. In the present embodiment, a plurality of directions to be extracted as a row of pixels that are continuous in each direction are four directions of vertical, horizontal, rising right, 45 degrees, and rising 45 degrees as shown in FIG. In the following description, these four directions are referred to as V, H, R, and L, respectively. In a practical application of the present invention, the extraction direction is not limited to these four directions.

First, the processing in the direction continuous value providing unit 201 will be described. The direction continuation value assigning unit 201 defines the direction continuation value assigned to each pixel as the number of consecutive black pixels when an image is scanned from one target pixel in both directions in each direction. For example, in the example shown in FIG. 4, when the image is scanned rightward and leftward in the H direction from the pixel of interest 403, six black pixels are continuous. 6 ". When the target pixel is a white pixel, the direction continuation value in each direction is “0”.

In the other direction continuous values of the target pixel 403, V is "3", R is "3", and L is "5".

The above direction continuous value used in the present embodiment is:
It can be obtained in a simple and parallelizable way. An example of extracting a direction continuous value in the H direction using this method will be described with reference to FIGS. 5 and 11. 5 and 11, P _{x, y} indicates the coordinate values of the pixels x, y of the input image, H _{x, y} indicates the continuous value in the horizontal direction, w indicates the pixel width, and h indicates the pixel height. First, in steps 10 and 11, initialization is performed by substituting y = 0 and x = 1. Next, in step 12, it is determined whether P _{0, y} is a black pixel. If it is not a black pixel, the process proceeds to step 13 and H _{0, y is set} to 0. If it is a black pixel, the process proceeds to step 14 and H _{0, y is set} to 1. Fig. 5 (a)
Since P _0,0 is white in the example of the pixel row shown in ( ₁₎ , the process proceeds to step 13, and H _1,0 becomes 0. Next, the routine proceeds to step 15, where it is determined whether or not P _{x, y} is a black pixel. If it is not a black pixel, the process proceeds to step 16 and H _{x, y is set} to 0, while if it is a black pixel, the horizontal continuous value H _{x-1, y} on the left is 1
Is added. In the example of the pixel column shown in FIG.
_{Since 1,0} is white, the process proceeds to step 16, and _H1,0 becomes 0. Next, the process proceeds to step 18 where x is incremented. In step 19, it is determined whether or not x is less than w (that is,
It is determined whether or not the scanning has been performed to the right end). If x <w, the process returns to step 15 until step x <w is not satisfied.
To 19 are repeated. As a result, one pixel row arranged in the H direction is scanned, and a primary horizontal continuous value H _{x, 0} is determined as shown in FIG. 5B. In step 19, when it is determined that x <w is not satisfied, the process proceeds to the next step 20, where x = w−2 is substituted. The reason why 2 is subtracted is that the final horizontal continuity value has already been obtained for the rightmost pixel 501. Next, the process proceeds to step 21, where it is determined whether the pixel of interest is not 0 (ie, black) and the pixel on the right of the pixel of interest is not 0 (ie, black). If the pixel of interest is black and the pixel on the right is black, the process proceeds to step 22, where the value assigned to the pixel of interest is made equal to the value assigned to the pixel on the right. For example, in the case of the target pixel 503, the value is equal to the value “2” of the pixel 592 on the right side. If the target pixel is white or the pixel on the right side is white, nothing is performed, and the process proceeds to step 23, where x = x−1. Or not). If x ≧ 0, step 15
And the processing of steps 21 to 24 is repeated until x ≧ 0 is not satisfied. By the one round trip scanning process, as shown in FIG. 5 (c), the continuation value in the H direction of all the pixels included in one pixel row arranged in the H direction is obtained at a time. When it is determined in step 24 that x ≧ 0 is not satisfied, the process proceeds to the next step 25, where y = y + 1, and the process proceeds to the next line. In step 26, it is determined whether y <h (that is, scanning is performed up to the last line). Is determined). y <
If h, the process returns to step 11, and if y <h, the process ends. The above processing can be performed independently of the other pixel rows arranged in the H direction, so that parallelization is possible.

The other continuity values in the V, R, and L directions can also be obtained by performing the same processing on one pixel row arranged in the V, R, and L directions.

The purpose of the processing in the direction continuous value diffusion unit is as follows.
The purpose is to correct the direction continuity value so that the direction attribute assigned to each pixel in the direction attribute determination unit is not adversely affected by fine irregularities in the contour portion. Therefore, in the actual process, the process in the direction continuous value diffusion unit is first, but in the description, the process in the direction attribute determination unit is first.

The direction attribute determination unit determines which direction attribute the pixel has from the direction continuity value for each direction, which is provided to each pixel by the direction continuity value provision unit and updated by the direction continuity value diffusion unit. Then, a direction attribute is given to the pixel. In this embodiment, four of V, H, R, and L are set for each pixel.
Direction continuation values for the directions are respectively given.

FIG. 12 shows a processing example of the direction attribute judging section. First, in steps 27 and 28, x = 0, y
= 0, and then proceed to steps 29 to 31, where the direction continuity values V ′ _{x, y} , H ′ _{x, y} ,
R ′ _{x, y} and L ′ _{x, y} are compared, and the direction giving the maximum direction continuous value in steps 32 to 35 is set as the direction attribute C _{x, y} to be given to the pixel. For example, the above noted pixel 4
Since the direction having the maximum direction continuous value at 03 is “H”, the direction attribute assigned to this pixel is “H”.
If there are a plurality of directions that give the maximum direction continuous value, all of those directions are given. However, since the continuous values in the R and L directions count pixels arranged at an angle of 45 degrees, it is necessary to evaluate the geometric length by multiplying the continuous values in the H and V directions by the square root of 2. Therefore, when comparing the above continuous direction values, the comparison is performed after multiplying the continuous direction values in the R and L directions by the square root of 2. Then, proceed to step 36, where y = y
Then, the process proceeds to the next line as +1. In step 26, it is determined whether y <h (that is, whether the scanning has been performed up to the last line). If y <h, return to step 29,
Steps 29 to 37 are repeated. If y <h, the process proceeds to the next step 38, where x = x + 1, and moves to the next column. In step 26, it is determined whether x <w (that is, whether scanning has been performed to the left end). x <w
If so, the process returns to step 28; otherwise, the process ends.

It is preferable that the pixel to which the direction attribute is assigned is a pixel at the outline of the character. The definition of the pixel of the contour portion is such that the pixel is black and at least one of four pixels adjacent to the upper, lower, left and right is white.

Next, the processing in the direction continuous diffusion section will be described. It is possible to use the direction continuation value obtained by the direction continuity providing unit as it is to assign a direction attribute by the direction attribute determination unit, but the direction attribute fluctuates due to fine irregularities in the contour part. There is a point. This problem will be described with reference to FIG. Taking the image as shown in FIG. 6 (a) as an example, the outline of such a horizontally long line
It is natural to give the direction attribute “H” in the sense of expressing the characteristic, and the direction continuous value in the H direction obtained by the direction continuous value providing unit for this image is shown in FIG. As for the direction continuous value in the H direction, a large value is obtained inside the line, but only a small value is obtained in the contour part, especially in the protruding part. The outline pixel indicated by a black circle in FIG. 6C does not have the direction attribute “H”, but has another direction attribute.

In order to compensate for this problem, it is effective to diffuse large continuity values obtained inside the line on the image, and to give large continuity values to pixels in the outline. FIG. 13 shows an example of processing by the direction continuous diffusion unit. First, at step 40, x = 0, and the process proceeds to the next step 41, where the direction continuous value H of the pixels in the first row obtained by the direction continuous value providing unit is set. _{The value x, 0} is compared with the direction continuation value H _{x, 1 in} the second row, and the largest value is reassigned as the direction continuation value H ′ _{x, 0} . Then, proceed to step 42, set y = 1,
The process further proceeds to step 43, where the direction continuation value H of the pixel of interest is obtained.
_{x, y} , the direction continuation value H _{x, y-1} of the row above it and the direction continuation value H _{x, y + 1} of the row below it are compared, and the largest value among them is the direction continuation value H ' _Reassign as _{x, y} . Then, the process proceeds to step 44, where y = y + 1. Further, the process proceeds to step 45, where it is determined whether y <h−1 (that is, whether scanning has been performed up to the row immediately before the last row). If y <h-1, the process returns to step 43, and the processes of steps 43 to 45 are repeated. On the other hand, if y <h−1, the process proceeds to the next step 46, where the direction continuous value H _{x, h−1 of} the last row is compared with the direction continuous value H _{x, h−2 of the} immediately preceding row. , Is re-assigned as the direction continuous value _{H'x, h-1} . Then, the process proceeds to step 47, where x = x + 1, and further proceeds to the next step 48, where it is determined whether or not x <w (that is, whether or not scanning has been performed to the left end). Returning to 41, if x <w, the process ends. Sixth
When the direction continuous value shown in FIG. 6B is updated by this processing, a direction continuous value as shown in FIG. 6D is obtained. For example, the contour pixel 60 of the protrusion shown in FIG.
In 5, the direction continuation value “2” is assigned before the update, but after the update, “13” is assigned to the pixel 605 ′ at the same coordinate position, and a sufficiently large direction continuation value is obtained. When the direction attribute is determined using the direction continuation value, all the pixels in the outline portion of this figure have the direction attribute “H”, and the global line direction can be accurately extracted without being influenced by the unevenness of the outline portion. Can be.

The same update processing is performed on the direction continuous values in the other V, R, and L directions. In this embodiment, V, H,
FIG. 7 shows the range of pixels referred to in the process of updating the direction continuation value for each of the R and L directions.

The area dividing section divides a space corresponding to an input image into elementary areas. In this embodiment, as shown in FIG. It is divided into the elementary regions.

Next, the processing in the direction attribute counting section will be described. The direction attribute counting unit counts pixels having each direction attribute in each elementary region. In this embodiment, the character image is divided into nine regions as shown in FIG.
Since pixels having four direction attributes of H, R, and L are counted, as shown in FIG. 9A, 9 × 4 = 36 types of values are obtained.

FIG. 14 shows a processing example of the direction attribute counting unit, where M _n is the number of the elementary region and F _n is the fifteenth region.
The data area shown in the figure and N represent the number of elementary areas (N = 9 in the example of FIG. 8). First, in step 49, n
= 0, and in the next steps 50 to 53, the elementary region M
Count pixels having each direction attribute V, H, R, L in _n ,
Data area _{F n, F n + N,} F n + 2N, if the values are the F _{n + 3N.} Next, the process proceeds to step 54, where n = n + 1, and further proceeds to the next step 55, where it is determined whether or not n <N. If n <N, the process returns to step 50, and step 50 is performed.
To 55 are repeated. If n <N, the process is terminated. As a result, as shown in FIG.
Data arranged for each of H, R, and L can be obtained.

In the example of FIG. 8, if the direction attribute is given to all the pixels existing in the outline of the image, the result is as shown in the example of FIG. 8 (c). FIG. 8 (b) shows the element region 8.
02 is an enlarged view of the image 81 of the elementary region.
Focusing on the peripheral pixel 811 of 0, if the direction attribute of this pixel is determined by the above-described process, this is “L”. By determining each direction attribute for all peripheral pixels of the image and counting pixels having direction attributes in each direction in the elementary region 802, “H” is 0 and “V” is 1
4, "R" is 3, and "L" is 5.

The direction attribute set is composed of the elementary regions 801 and 80
The results of counting the pixels of each direction attribute in 2, 803,... 809 are collected in a unit area 800 for one character to create the pattern shown in FIG. 9B. This is an image feature pattern extracted by processing the read characters and symbols.

The dictionary pattern storage device 104 is a memory in which image features are extracted from standard characters in advance by the same means as described above and image feature patterns of these standard characters are stored.

The feature comparison unit 105 is provided with the direction attribute collection unit 20
6, the image feature pattern of the read image output from the dictionary pattern storage device 104 is compared with the image feature pattern of the standard character sequentially read from the dictionary pattern storage device 104. The most similar of both patterns is the recognized character or symbol.

As an example of means for determining the degree of similarity between the two patterns, there is a method of calculating the Euclidean distance between the two image feature patterns. Assuming that both patterns are shown in FIG. 10, the Euclidean distance between them is as follows.

[0036]

(Equation 1)

This recognition result is obtained by using a character code (for example, ASC) defined in a system in which this device is incorporated.
II) and output from the character recognition device.

In this embodiment, the case where the image feature extraction device included in the present invention is used for recognition of characters and the like has been described. Widely applicable to separation and the like.

In this embodiment, the definition of the direction continuity value is the number of consecutive black pixels when the image is scanned in both directions from the target pixel in each direction. However, another definition may be used.
For example, when a small number of white pixels appear in the middle of scanning and black pixels continue again, counting can be continued to make it less susceptible to noise. Alternatively, the number of black pixels that appear when scanning within a certain range in both directions from the target pixel in each direction may be used. Further, in the embodiment, the direction continuous value “0” is assigned to the white pixel, but the number of consecutive white pixels when the image is scanned in both directions from the target pixel is set as the direction continuous value. Accordingly, it is also possible to assign a continuous direction value other than “0” to the white pixel. In this embodiment, a binary image is targeted. However, a multi-valued image is targeted, and a sum of densities when scanning within a certain range in both directions from a target pixel in each direction may be used as a direction continuous value. Good.

In the present embodiment, in the processing in the direction continuous value diffusion section 202, the maximum value of the direction continuous values of the pixels near the target image is used, but the average value of the nearby pixels may be used instead.

In the present embodiment, when there are a plurality of directions giving the maximum direction attribute in the processing of the direction continuity determining unit 203, all of these directions are given as direction attributes.
Only one of the directions may be given as the direction attribute.

In the present embodiment, the definition of the pixel of the outline portion is such that the pixel is black and at least one of the four adjacent pixels is white. The pixel may be such that at least one of the eight adjacent pixels on the left and right is white. Further, in the present embodiment, the direction attribute is given only to the pixels of the contour portion, but the direction attribute may be given to all the black pixels. According to the method of giving directionality only to the outline portion, even if the thickness of the character line changes, the change in the number of pixels having the direction attribute is small, and the image feature can be stably extracted.

Alternatively, the skeleton of the image may be extracted and a direction attribute may be assigned to pixels on the skeleton. Further, the direction attribute value may be given to the white pixel, and then the direction attribute may be given to the white pixel.

In this embodiment, the unit area is decomposed into nine elementary areas, but the number of decompositions is not limited to nine as described above. In addition, the respective element regions may overlap. Further, the counting of pixels having the direction attribute for each direction may be omitted for all directions in the elementary region, and may be performed for only a part of the direction attributes according to the position of the elementary region.

In the present embodiment, the Euclidean distance is used as a measure of the degree of similarity between patterns in the feature comparison device 105, but another measure may be used. For example, Mahalanobis distance, simple similarity, composite similarity, and the like are applicable.

[0046]

As described in detail above, the character recognition apparatus of the present invention performs a process of assigning a direction continuous value to each pixel of an image and diffusing the direction continuous value. As a result, adverse effects from noise such as fine irregularities in the outline of the character and dirt in the target area are removed. The data to be handled is an integer, and the process according to this configuration can be performed in parallel, so that the entire process is speeded up.
Further, this configuration is simpler than the conventional technology. Therefore, it does not take long time to extract features such as characters, the literacy rate is improved, and the apparatus can be provided economically.

[0047]

[Brief description of the drawings]

FIG. 1 is an overall configuration diagram of characters and the like according to the present invention;

FIG. 2 is a configuration diagram showing a configuration of a feature extraction device provided in the present invention;

FIG. 3 is an explanatory view in four directions used in the embodiment;

FIG. 4 is an explanatory diagram of a direction continuous value,

FIG. 5 is an explanatory diagram of a method of extracting direction continuous values,

FIG. 6 is an explanatory diagram of a process of a direction continuous value spreading unit,

FIG. 7 is an explanatory diagram showing a pixel range referred to in a process of updating a direction continuous value in each direction in the embodiment.

FIG. 8 is an explanatory diagram showing an example of an elementary region divided by a region dividing unit;

FIG. 9 is an explanatory diagram showing an image feature pattern;

FIG. 10 is an explanatory diagram showing symbols of a Euclidean distance calculation formula,

FIG. 11 is a flowchart illustrating an example of a process of extracting continuous direction values in the horizontal direction.

FIG. 12 is a flowchart illustrating an example of a process of assigning a direction attribute,

FIG. 13 is a flowchart showing an example of a process of diffusing a horizontal direction continuous value;

FIG. 14 is a flowchart illustrating an example of a process of calculating a feature amount by counting direction attributes;

FIG. 15 is an explanatory diagram showing a data area for storing counted direction attributes.

[Description of Signs] 101: Image input device, 102: Pre-processing device, 103:
Feature extraction device, 104: dictionary pattern storage device, 10
5: feature comparison device, 201: direction continuous value giving unit, 20
2: Direction continuous value diffusion unit, 204: region division unit, 205:
Direction attribute counting unit, 206: Direction attribute collection unit, 301: pixel, 401, 501, 601: white pixel, 402, 50
2, 602: black pixel, 403, 701, 702, 70
3, 704, 811: pixel of interest, 603: direction continuous value in the H direction obtained by the direction continuous value assigning unit, 604: pixel of a contour part whose direction attribute is "H", 605: direction attribute other than "H" , 606: H-direction continuity value obtained by the direction continuity value diffusion unit, 800: unit area for one character, 801, 802,... 809: elementary area.

────────────────────────────────────────────────── (5) Continuation of the front page (56) References JP-A-3-175591 (JP, A) JP-A-3-182986 (JP, A) JP-A-59-184977 (JP, A) JP-A-63-184977 157290 (JP, A) JP-A-2-15388 (JP, A) JP-A-2-228788 (JP, A) JP-A-2-18784 (JP, A) JP-A-56-87169 (JP, A) (58) Field surveyed (Int.Cl. ⁷ , DB name) G06K 9/46 G06K 9/48 G06K 9/62 G06T 7/60 JICST file (JOIS)

Claims (4)

(57) [Claims] 1. A direction continuity value assigning means for examining continuity in a plurality of directions with respect to each pixel of an input image, and assigning a direction continuity value representing the continuity to each direction. The direction continuity value of each pixel obtained by the assigning means is subjected to a process of diffusing the direction continuity value of the pixel inside the contour to the direction continuation value of the contour pixel in a space corresponding to the input image, so that the direction continuity value is obtained. A direction continuous value diffusing unit to be changed, a direction attribute determining unit that determines which direction attribute the pixel of interest has based on the direction continuous value of each pixel changed by the direction continuous value diffusing unit, and the direction attribute determining unit A direction attribute counting unit for counting pixels having the direction attribute determined by the above for each direction attribute. 2. A direction continuity value assigning means for examining continuity in a plurality of directions for each pixel of an input image, and assigning a direction continuity value representing the continuity to each direction. The direction continuity value of each pixel obtained by the assigning means is subjected to a process of diffusing the direction continuity value of the pixel inside the contour to the direction continuation value of the contour pixel in a space corresponding to the input image, so that the direction continuity value is obtained. A direction continuous value diffusion unit to be changed; a direction attribute determination unit for determining which direction attribute the target pixel has based on the direction continuous value of each pixel changed by the direction continuous value diffusion unit; Area dividing means for dividing the space into a plurality of areas; and pixels having the direction attribute determined by the direction attribute determining means in each of the areas divided by the area dividing means. A feature extraction device comprising: a direction attribute counting means for counting. 3. A step 1 in which continuity in a plurality of directions is checked for each pixel of the input image, and a direction continuity value representing the continuity is given to each direction. The direction continuous value of each pixel is
Step 2 of changing the direction continuity value by performing a process of diffusing the direction continuity value of the pixel inside the contour into the direction continuation value of the contour pixel in the space corresponding to the input image; A step of determining which direction attribute the target pixel has based on the direction continuous value of the pixel; and a step of counting pixels having the direction attribute determined in step 3 for each direction attribute. The feature extraction method to be the feature. 4. A step 1 in which continuity in a plurality of directions is checked for each pixel of the input image, and a direction continuity value representing the continuity is given to each direction. The direction continuous value of each pixel is
Step 2 of changing the direction continuity value by performing a process of diffusing the direction continuity value of the pixel inside the contour into the direction continuation value of the contour pixel in the space corresponding to the input image; Step 3 of determining which direction attribute the pixel of interest has based on the direction continuation value of the pixel; Step 4 of dividing the space corresponding to the input image into a plurality of regions; A step of counting pixels having the direction attribute determined in step 3 for each direction attribute.