JP2635779B2 - Drawing feature extraction device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for extracting various image features from an image in which various image information such as drawings, particularly topographic maps, are superimposed and included.

[Conventional technology]

Conventionally, various proposals have been made for this type of extraction, but all of them usually depend on dedicated hardware or software.

For visual inspection and the like, those that perform non-directional contraction / expansion calculations have been put to practical use.

[Problems to be solved by the invention]

However, the former has a problem that processing contents are limited and lacks versatility, and the latter has a problem that it takes time.

Therefore, an object of the present invention is to enable high-speed processing while having versatility.

[Means for solving the problem]

An image memory that stores an image of a drawing imaged by an imaging device, a working memory that stores an intermediate result of processing as a directional image, a result storage memory that stores a final result of processing, the image memory or the working memory A knot that performs [not (knot) operation and bit rotation (rotate) operation] or [only bit rotation (rotate) operation] of image data read from the memory
A rotate circuit, a local data generation circuit for generating local data from the output thereof, and a negation (knot) operation of the image data read from the image memory or the working memory, or the data as it is without performing any operation A knot circuit for outputting, a local mask memory for storing various local masks, a local mask selecting circuit for selecting a local mask from the local mask memory, an output of the local data creating circuit and an output of the local mask selecting circuit. Unary AND circuit and unary OR circuit that respectively perform AND operation and OR operation, and binary AND circuit and binary OR that perform AND operation and OR operation of the output of the local data creation circuit and the output value of the knot circuit, respectively A working memory or a result storage memo for the circuit and the outputs of the unary AND circuit, the unary OR circuit, the binomial AND circuit, And a memory writing circuit for writing the geometrical characteristics of the drawing by repeatedly performing a local parallel operation for each direction according to a predetermined sequence for each of the direction plane images obtained from the density gradient of the input image. .

[Action]

The point-and-point information and point-by-point information can be obtained by realizing the expansion / contraction operation taking into account the directionality by using a combination of 8-parallel unary AND circuit, unary OR circuit, binomial AND circuit, and binomial OR circuit using local masks Various image information including linear information can be obtained at high speed by local parallel operation.

 Here, the principle will be described.

I) Basic operator of directional contraction / expansion and feature extraction First, an image taken from a scanner or the like is binarized, and a feature field in eight directions according to the direction of a contour pixel (edge) as shown in FIG. (B) to (i)) are obtained. The eight feature fields are represented as a binary image in which the position where the edge in each direction is detected is “1”, and the other positions are “0”. The direction of the edge is the normal direction from the black (white) pixel to the white (black) pixel. FIG. 3 shows an example of a directional feature field obtained for a 1: 25,000 topographic map. FIG. 7A shows an original image, and FIG. 7B shows an extraction result of the directional feature. The present invention is intended to extract image information of roads, buildings, hatched areas, and the like in a locally parallel manner by combining repetitive application of basic operations mainly for contraction and expansion to the eight-direction plane image. . FIG. 4 shows a list of types of basic operators.

Among the basic operators shown in FIG. 4, d direction of the fan-shaped contraction operator S _'d is b a binary image, b [d] a d direction (d = 0
If the shift images to _{~7), S 'd: S} ' d b = b∩ (b [d-1] ∪b [d] ∪b [d
+1])... (1) The contraction is performed in a fan shape in order to erode (delete) only one pixel at the end point of the line segment in the direction plane. The operator S _d for shrinking the line segment of the d-th surface in a fan shape from both directions can be expressed as S _d : S ′ _{d + 2} ∩S ′ _{d + 6} (2) using S _d ′. That is, the fan-shaped contraction calculation is performed from a direction different from the d direction by 90 degrees. Unidirectional dilation operator L ' _db = b
∪b [d] (3) X ′ _d : x ′ _d b = b∪ (b [d−1] ∪b [d] ∪b [d
+1])... (4) Operators x _d of the fan-shaped expanding a segment of the d surface in both the 'with _{d, x d: x' X} d + 2 ∪X 'd + 6 ...... (5) represented. The directional dilation operation M _f with a mask is as follows: f is a mask image, D ₈ is a non-directional dilation operator near ₈ and M _f : M _f b = _f bD ₈ b (6) D ₈ : D ₈ b = b∪b [0] ∪∪b [1] ∪b [2] ∪b
[3] ∪b [4] ∪b [5] ∪b [6] ∪b [7]. This is after performing directional contraction,
It is used to restore the original image in the dilation operation.

Various types of image information are obtained by repeatedly applying these basic operators and combinations thereof. Hereinafter, a description will be given of an example of extracting a linear feature in which the characteristics of the directional contraction / expansion calculation are well exhibited.

1) First, the image ▲ f ^short _d removing the following line 6 pixels
▼ (also written as f (short, d)).

It should be ^{_{noted, ▲ M 3 ad S 3 d}} ▼ a d is also referred to as _{M (3, a d) S} (3, d) a d. Here, the first number of the operator indicates the number of repetitions, the letter a _d indicates the original directional surface image, and d indicates the direction.
When two operators are arranged, the operation is performed in order from the right. Thus, equation (7) after performing three times the operation of one pixel at a time removed from both ends of a line segment of the direction plane image a _d, 6 pixels or less inflated 3 times the original direction plane image a _d as a mask image Represents a process of removing the line segment. This is a major feature of the directional expansion / contraction calculation.
In the non-directional contraction calculation, both "points" and "lines" disappear, but by performing contraction and expansion calculations from both directions of the line segment on the directional feature plane, "points" disappear and "lines" disappear. Can be left.

2) Obtain the end point ▲ f ^short _d ▼ (also referred to as f (end, d)).

3) Line ^{し て} f ^conect _d ▼ (f (cone
ct, d)).

4) Relax the direction accuracy.

5) Extract long lines.

In addition, various pieces of image information such as dot information, hatched areas, and roads drawn with parallel lines can be similarly extracted by combining repeated application of the basic operator.

II) Cyclic processing using a local mask and four AND and OR circuits The above-mentioned repetitive application of the basic operators and the extraction of image features by their combination are performed as follows: (a) Image data creation circuit for operation, negation of read image data A circuit for performing a bit rotation process by circulating data taken out by a shift process without erasing, a circuit for generating 3 × 3 local data from an output thereof and outputting the data to an internal input data bus I, and a read image A circuit for negating the data and outputting it to the internal input data bus II, (b) a circuit for selecting various local masks for directional contraction and expansion as an auxiliary circuit for the operation, and (c) an internal data bus I as an arithmetic circuit
A unary OR circuit that performs an OR operation using the above data and a local mask, a unary AND circuit that performs an AND operation, and performs an OR operation between the center pixel on the previous internal input data bus I and the pixel on the previous internal input data bus II A binary OR circuit, a binary AND circuit that performs an AND operation, and can be obtained by the cyclic processing of the above circuit group.

Now, let the input images of the four circuits be P = (p _ij ) and Q =
(Q _ij ), the output image is R = (r _ij ), (i = 1 to n, j = 1
To n), the local mask is M = (m _kl ), (k = 1 to 3, l = 1
３3), the unary OR circuit provides one input image P Unary AND circuits are similarly The binary OR circuit has two input images P and Q, r _ij = p _ij ∪q _ij ... (14) The binary AND circuit similarly has r _ij = p _ij ∩q _ij ... (15) Are respectively performed. The operations of the four circuits are all performed in eight parallels.

The fan-shaped contraction operation S ′ _d is first performed by using a local mask m as shown in FIG.
Using S'0 to mS'7, a sector-like shift is performed by a unary OR circuit. Next, the shifted image and the original image are read to two data buses, and a fan-shaped contraction operation is performed by a binary AND circuit.

Sector shrinkage operation from both directions S ' _d : S' _{d + 2} ∩S '
_{d + 6 is, S d: S d b =} (b∩ (b [d + 1] ∪b [d + 2] ∪b [d + 3])) ∩ (b∩ (b [d + 5] ∪b [d + 6] ∪b [d + 7 ])) = B∩ (b [d + 1] ∪b [d + 2] ∪b [d + 3]) ∩ (b [d + 5] ∪b [d + 6] ∪ (b [d + 7]) (16) Local masks mS'0 to mS'7 shown in FIG.
After performing a sector shift in the d + 2 direction and the d + 6 direction by the unary OR circuit using, the logical product of them is obtained by the binary AND circuit, and the logical product of the image and the original image is obtained again by the binary AND circuit. This is achieved by:

Fan dilations X _'d are determined in one unary OR operation using the local mask mX'0~mX'7 of FIG. 6 (a) ~ (h).

The sector expansion operation X _d : X ′ _{d + 2} ∪X ′ _{d + 6} in both directions is performed by one unary OR operation using the four local masks mX04 to mX37 shown in FIGS. 7 (a) to 7 (d). Desired.

The unidirectional expansion operation L' _d is shown in FIGS. 8 (a) to 8 (h).
Is obtained by one unary OR operation using the local masks mL'0 to mL'7.

Unidirectional dilation operation in both directions L _d : L ′ _{d + 2} ∪
L ′ _{d + 6} is the local mask mL04 to mL3 in FIGS. 9 (a) to 9 (d).
It can be obtained by one unary OR operation using 7.

The unidirectional contraction operation C is the same as the operation L' _{d in} FIG.
(H) can be obtained by one unary AND operation using the local masks mL′0 to mL′7.

The non-directional expansion operation D ₄ near ₄ and the non-directional expansion operation D 8 near ₈ are the local mask mD in FIGS. 10 (a) and (b), respectively.
4, determined by one unary OR operation using mD8.

The non-directional contraction operation E ₄ near ₄ and the non-directional contraction operation E 8 near ₈ are the same as those in the dilation operation shown in FIGS.
It is obtained by one unary AND operation using the local masks mD4 and mD8 in (b).

Note that the non-directional contraction operation E is expressed as follows: E: Eb = b {b [0] {b [1] {b [2] {b [3]}
b [4] ∩b [5] ∩b6] ∩b [7].

The non-directional dilation operation M _f and the directional dilation operation W _{d: f} using a certain image f as a mask are performed by performing a dilation operation with a unary OR circuit using the corresponding local mask, and then performing the dilation operation between the image and the mask image f. The logical product is obtained by a binary AND circuit.

The isolated point removal operation Z performs the
After performing a unary AND operation with the local mask mZ in FIG. 11,
It can be obtained by a binary AND operation between it and the original image.

The basic operator is a unary OR using a local mask as described above.
It is executed by a circuit, a unary AND circuit, a binary OR and a binary AND circuit. Since various types of image information can be obtained by a combination of repeated application of basic operators and a logical product or logical sum between intermediate result images, a unary OR circuit using a local mask,
It can be obtained by the cyclic processing of the unary AND circuit, the binary OR circuit, and the binary AND circuit.

〔Example〕

FIG. 1 is a block diagram showing an embodiment of the present invention. In FIG. 1, reference numeral 1 denotes an image (image) memory for storing an image of a drawing imaged by an imaging device; 14, a working memory for storing an in-process result as a direction plane image;
5 is a memory readout circuit, 15 is a result storage memory for storing the final result of the processing, 3 is [Negation (knot) operation and bit rotation (rotate) operation] of image data read from the image memory or the working memory or A not-rotate circuit for performing [bit rotation (rotate) operation only], a local data generating circuit 4 for generating local data from its output, 6 a negation (not) of image data read from the image memory or the working memory. A knot circuit that outputs data without performing any operation or without performing any operation, 7 is a local mask memory that stores various local masks, 8 is a local mask selection circuit that selects a local mask from this local mask memory, 9 and 10 are the outputs of the local data creation circuit and the local mask selection circuit. AND operation with the output, a unary AND circuit and an unary OR circuit, which respectively perform the OR operation, 11 and 12 are the output of the local data creation circuit and the output value of the knot circuit.
Binary AND circuit that performs AND operation and OR operation, respectively, and binary O
An R circuit 13 is a memory writing circuit for writing the outputs of the unary AND circuit, unary OR circuit, binomial AND circuit, and binomial OR circuit into a working memory or a result storage memory.

Hereinafter, extraction of linear information, roads, buildings, dots, and hatched areas will be described in order.

(1) Extraction of linear information (1-1) ▲ f ^short _dに shown in the above equation (7) is calculated.

▲ f ^short _d ▼ is obtained by performing three masked sector contractions and three dilation operations. In this fan-shaped contraction operation, as described above, the logical product between the images sector-shifted in the d + 2 direction and the d + 6 direction may be obtained, and the logical product of this and the original image may be obtained. Therefore, first, d + 2 by the unary OR operation
Perform a sector shift in the direction.

Hereinafter, description will be made with reference to FIG. As shown in FIG. 12, the image memory 1 stores the data of the eight-direction feature plane in a byte-packed form. The memory reading circuit 2 reads 1-byte image data a from the image memory 1 and sends the image data b to the NOT / ROTATE circuit 3. The NOT · ROTATE circuit 3, if necessary negative data (inversion) and performs inter-plane Roteito but, ▲ f ^short _d
In the case of ▼, nothing is performed, and the image data c is sent to the next 3 × 3 local data creation circuit 4. The 3 × 3 local data creation circuit 4 first uses the internal two-dimensional local memory to
× 3 byte-type local data is created, and as shown in FIG. 13, 8 × 3 × 3 bit-type local data d1 to d8
Output to the input data bus I.

On the other hand, the local mask selection circuit 8 stores the local mask m for eight planes shown in FIGS.
S'0 to mS'7 are selected and sent to the 8-parallel unary OR circuits 10a to 10h as local mask data h1 to h8. In this case, since the contraction is performed from one of the line segments, the local mask on the d-plane is
mS '(d + 2). The unary OR circuits 10a to 10h perform unary OR operations on the 3 × 3 bit-type local data d1 to d8 by the local mask data h1 to h8, respectively. On the output data bus, the 1-bit operation result of the circuit is handled as 1-byte operation result data i for 8 surfaces. The memory writing circuit 13 writes the operation result i into the working memory 14a.

By the above processing, the image shifted fanwise in the d + 2 direction is stored in the working memory 14a. By exactly the same processing, an image shifted sectorwise in the d + 6 direction is obtained and stored in the working memory 14b. The difference at this time is
The only point is that the local mask data on the d-plane is mS '(d + 6).

Next, an AND operation is performed on the sector shift image in the d + 2 direction and the sector shift image in the d + 6 direction. Similarly to the above, the memory readout circuit 2, the NOT ROTATE circuit 3, and the 3 × 3 local data creation circuit 4 read out the d + 2-direction sector-shift image data from the working memory 14a to the data bus I as 3 × 3 local data d1 to d8. . On the other hand, the memory readout circuit 5 reads out the d + 6 fan-shaped shift image data from the working memory 14b and sends it to the NOT circuit 6. In this case, the NOT circuit 6 does not perform the negation operation, and the bit data f1 to f8
To the data bus II. 8 parallel binary AND
The circuits 11a to 11h are provided with a binary AN with a pixel on the input data bus II.
A D operation is performed, and operation result data i is output to an output data bus. The memory writing circuit 13 writes the operation result data i into the working memory 14c.

The AND data between the fan-shifted images has been obtained as described above. Next, an AND operation with the original image is performed. This can be performed in the same manner as in the above-mentioned binary AND operation. The input is the image memory 1 and the working memory 14c, and the output is the working memory 14a.

With the processing so far, one fan-shaped contraction calculation _Sd is completed. Furthermore, for the image S _d a _d of the working memory 14a,
Perform the same operation. At this time, the results of the two unary OR operations are temporarily stored in the working memories 14b and 14c, and the results of the binary AND operation are temporarily stored in the working memory 14d. Finally, the results of two fan-shaped contraction operations SS ² _d ▼ a _d Is stored in the working memory 14b. Similarly, three fan-shaped contraction operation result ▲ S ³ _{▼ d} a _d is obtained in the working memory 14c.

Next, a non-directional expansion operation _Mad using a _d as a mask is applied to this image three times. After performing the unary OR operation on the target image, the process of performing the binary AND operation with a _d is the same as the above-described operation. However, in the case of the non-directional expansion operation, the process shown in FIG. The local mask mD8 is used. Calculation result ▲ f
^short _d ▼ is stored in the operation result memory 15a. FIG. 14A shows an example of extracting ff ^short _dし^た by adding eight faces.

(1-2) An ^end point ▲ f ^end _d ▼ is obtained from the above equation (8).

This is, as is clear from equation (8), by performing a directional fan-shaped contraction operation on ▲ f ^short _d ▼ and calculating the negation data and ▲ f ^short d
Performs a binary AND operation with _d ▼. At this time, the NOT / ROTATE circuit 3 negates the image data b.

(1-3) ▲ f ^conenct _d ^{示 す} shown in the above equation (9) is obtained.

Here, ▲ f ^end _d演算 is used as an original image to perform three unidirectional dilation operations L _d , and a binary OR operation of this and ff ^short _d ▲ is performed. As a local mask, mL'0 to mL'7 in Fig. 8
Is used.

(1-4) ▲ f ^relax _d ▼ shown in the above equation (10) is obtained.

The even-numbered plane of ff ^conenct _d ^{求 め} obtained in the above item (1-3) can be obtained by an OR operation between adjacent planes.
^conenct _d ▼ is read by two memory read circuits 2 and 5,
The data of the input data bus I system is bit-rotated by the NOT / ROTATE circuit 3, and then a binary OR operation is performed.

(1-5) ▲ f ^long _d示^す shown in the above equation (11) is obtained.

Here, the fan-shaped contraction is performed 20 times and the 8-direction expansion calculation with mask is performed 17 times for the ff ^relax _d ^得 obtained in the section (1-4).

(2) Extraction of road drawn with parallel lines The road drawn with parallel lines is extracted by the following procedure.

(1) Perform a binary OR operation on ▲ f ^long _d ▼ and ▲ f ^conenct _d ▼ obtained in the item.

For ▲ f ^narrow _d ▼, an expansion operation is performed using as a mask image. b represents an original image in an eight-direction plane.

^演算 f ^proad _d AND AND operation between opposing surfaces. Input data bus I
Bit data is rotated four times by the NOT / ROTATE circuit 3 to perform a binary AND operation.

With the above processing, the road drawn by the parallel line is extracted. FIG. 14 (b) shows an example of extraction.

(3) Building extraction Buildings are extracted according to the following procedure.

The contraction operation near 4 is performed by a unary AND circuit using the local mask mD4 in FIG. 10 (a).

Since the intersection of characters may remain in ▲ f ^block _d ▼, the building is extracted by the above operation. In the isolated point removal operation Z, first, a NOT AND ROTATE circuit 3 performs a unary AND operation using the local mask mZ shown in FIG. Ask for. An example of the extraction is shown in FIG.

(4) Dot extraction Dots are extracted by the following procedure.

A line segment of 4 pixels or less in the direction feature plane is removed by two vibration-type contraction / expansion calculations, an eight-direction feature plane is added, and then a line drawing area is filled by four masked dilation calculations.

An example of the extraction is shown in FIG.

(5) Extraction of hatched area A hatched area is extracted according to the following procedure.

The hatching, which is a diagonal line inclined to the right, exists on the third surface and the seventh surface of the directional feature surface.

First, diagonal lines are connected by two non-directional expansion calculations, and unnecessary diagonal lines other than hatched portions are removed by six non-directional contractions. Calculation of D ₈ unary OR circuit using a local mask mD8 of view the 10 (b), the calculation of E ₈ respectively obtained by unary AND circuit using the same local mask. FIG. 16 (b) shows an example of extraction.

〔The invention's effect〕

According to the present invention, the directional erosion / expansion operation can be executed at high speed by cyclic processing by simple four circuits of a unary OR circuit, a unary AND circuit, a binomial OR circuit, and a binomial AND circuit using a local mask. And the advantage that various image features can be extracted at high speed can be obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is an explanatory diagram of an edge direction and eight feature fields, FIG. 3 is an explanatory diagram for explaining an original image and its directional feature fields, and FIG. Is an explanatory diagram for explaining the types of basic operators of the erosion and expansion calculation in consideration of directionality, FIG. 5 is a configuration diagram showing a local mask for a fan-shaped expansion operation, and FIG. 6 is a local mask for a fan-shaped expansion operation. FIG. 7 is a block diagram showing a local mask for sector expansion calculation in both directions, FIG. 8 is a block diagram showing a local mask for unidirectional expansion calculation in both directions, and FIG. 9 is a unidirectional expansion calculation for both sides. FIG. 10 is a configuration diagram illustrating a local mask for non-directional expansion calculation, FIG. 11 is a configuration diagram illustrating a local mask for isolated point removal, and FIG. 12 is a diagram illustrating image data. , FIG. 13 is an explanatory diagram for describing 3 × 3 local data, and FIG. 14 is f (short, d). FIG. 15 is an explanatory diagram for explaining the extracted buildings and dots, and FIG. 16 is an explanatory diagram for explaining an example of hatching area extraction. DESCRIPTION OF SYMBOLS 1 ... Image memory, 2,5 ... Memory readout circuit, 3 ... NOT / ROTATE circuit, 4 ... 3x3 local data creation circuit, 6 ... Not (NOT) circuit, 7 ... Local mask Memory, 8: Local mask selection circuit, 9 (9a to 9h)
... Unary AND circuit, 10 (10a to 10h) ... Unary OR circuit, 11 (11a
~ 11h) ... Binary AND circuit, 12 (12a ~ 12h) ... Binary OR circuit,
13: memory writing circuit, 14 (14a to 14h): working memory, 15 (15a to 15h): result storage memory.

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Kazuhiko Yamamoto 1-1-4 Umezono, Tsukuba City, Ibaraki Prefecture Inside the Electronic Technology Research Institute (72) Inventor Taiichi Saito 1-1-4 Umezono Tsukuba City, Ibaraki Prefecture Inside the Electronic Technology Research Institute (72) Inventor Yasuo Hongo 1st Fujimachi, Hino-shi, Tokyo Fujifacom Control Co., Ltd. (72) Inventor Masanobu Nakamura 1st Fuji-cho, Hino-shi, Tokyo Fujifacom Control Ltd. (72) Inventor Shinji Matsui 1 Fuji-cho, Hino-shi, Tokyo Fujitacom Control Co., Ltd. Examiner Furu Yasuda (56) References Information Processing Society of Japan 39th (late 1989) National Convention Lecture Papers (▲ II ▼) 3K-6, 1989.10.16 884-885 "Basic hardware for MAP method expansion / contraction calculation" Nakamura et al. IEICE Technical Report PR U88-76 9-16, 1988 "Feature extraction of topographic maps using binary image directional local parallel computation" Matsui et al. IEICE Technical Report PR U88-66 49-56, 1988 "Local parallel operation of binary image in directional feature field" Yamada et al.

Claims (1)

(57) [Claims] An image memory for storing an image of a drawing imaged by an imaging device; a working memory for storing an intermediate result of the processing as a directional image; a result storage memory for storing a final result of the processing; A knot / rotate circuit for performing [not (knot) operation and bit rotation (rotate) operation] or [bit rotation (rotate) operation only] of image data read from the image memory or the working memory; A local data creation circuit to be created; a knot circuit that performs a negation (knot) operation on image data read from the image memory or the working memory or outputs data as it is without performing any operation; From a local mask memory that stores A local mask selection circuit that selects a local mask; a unary AND circuit and a unary OR circuit that respectively perform an AND operation and an OR operation of an output of the local data generation circuit and an output of the local mask selection circuit; and the local data generation circuit. AND circuit and a binary OR circuit for performing AND operation and OR operation of the output of the knot circuit and the output value of the knot circuit, respectively, and the outputs of the unary AND circuit, the unary OR circuit, the binary AND circuit, and the binary OR circuit And a memory writing circuit for writing the data in the working memory or the result storage memory.The local parallel operation in each direction is repeatedly performed in accordance with a predetermined sequence on each of the direction plane images obtained from the density gradient of the input image, so that the drawing of FIG. A drawing feature extracting apparatus for extracting a geometric feature.