JPS61208179A - Image describing device - Google Patents

Abstract

PURPOSE:To obtain a description system which is small in the amount of data by forming a hierachical structure of areas extracted by detecting adjacency and inclusion relation between areas, approximating boundaries and straight lines of the areas by a primitive such as an arc, and describing even an area which includes an island by an external approximate boundary. CONSTITUTION:An original image 8 is converted by an image input means 1 into a many-valued or polychromatic digital image to generate input image data 9; and a boundary extracting means 2 extracts a boundary point sequence of each area of the input image data, and an area hierachical structuring means 3 detects adjacency and inclusion relation between areas to generate extracted area data 10 consisting of hierachical data of areas and boundary point sequences. Then, a boundary approximating means 4 approximates the boundary point sequence of the extracted area data 10 by a straight-light primitive or arc primitive to generate approximate area data 11, so that a description data display means 6 displays the contents of image description data 12 on a CRT display device 7.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention provides a method for transmitting, storing, or processing images.
The present invention relates to an image description device that describes the boundaries of connected regions (hereinafter simply referred to as regions) constituting an image using primitives such as straight lines and circular arcs.

[Conventional technology]

Conventionally, as this type of image description device, the method described below (Reference ■: IB Ichiban M Journal
BM I, RB8, DEVELOP, Vow,
core A44DT).

7J4'~riJ'l November/Detco)
Reference) Figure 811 is a system configuration diagram of a conventional example, and the first
Figure 1 is a flowchart showing the procedure of the conventional example, and the first
FIG. 3 is a diagram illustrating the outline of the area.

In Figure 1/, (, zi) is an image input device, (λko) is an image memory, (yu3) is a processing program, (koda)
is the CPU (central processing unit), (X, t) is the working memory, (26) is the memory for the extraction area, and (kot) is the [1f
Ii is a memory for description, and (ri) is a CRT (cathode Nv) display. In addition, in FIG. 12, (9) is input image 1 image data, (2tz) is ko-subdued image data, (ko/
A) is contour point sequence data, and (/) is image description data.

As shown in Figure 1, in the conventional example, first step (2ti)
), input from the image input device @(ko/) is input to the image memory (
Multi-plant or multi-colored image data (
9) into predetermined gradation or color units according to the processing program (23), and converts each unit into direct image data (cotl). For example, black and white t The bit input ulIiill is classified according to two predetermined thresholds, and 1
Direct 1fii image data consisting of a set of pixels in one gradation range and a set of pixels in another gradation range (for example, convert the former to 'l' and the latter to 'O'). Create 3 sheets.

Next, in step (co/co), step (211)
The outline point sequence of the region to be extracted (for example, the region having the value "l") of the image data of the value "lj" obtained in the above is extracted as the boundary of the region. The contour point sequence here refers to a set of pixels located at the outermost position of the area among the pixels belonging to the area (outer contour point sequence), as shown in FIG.
In the case of an area with an island, a set of pixels surrounding the island (inner contour point sequence). 3×
This is done using 3 masks.

Finally, in step (C/J), the contour point sequence data (C/6) obtained in step C) is input,
The image is described by approximating the contour point sequence with straight line primitives. This approximation approximates a sequence of contour points using the vertex coordinates of a polygonal line within a predetermined tolerance.
A large number of methods have been devised so far (Reference ■, Reference ■: Magazine I.E.E.
ocjth.

Int, Conf, Pattern Recopn,
PP, tr3~trsq, t9
g:1.

IgFE). The composition of the image description data (I2) to be created is that each area has 7? gradations or colors. Each region is described by one approximate contour corresponding to the outer contour and, if islands are included, approximate contours corresponding to the number of islands corresponding to the inner contour.

In the conventional example, as described above, a region including an island is described by an approximate contour corresponding to an outer contour and an approximate contour corresponding to an inner contour, and an island region is described by an outer contour that is in contact with the inner contour of the region including an island. Since it is described by an approximate contour corresponding to
For areas that are islands, boundaries are defined in multiple ways.

[Problem 5 to be solved by the invention 1 As explained above, the conventional method separates a multi-valued or multi-colored image into gradation or color units, and calculates the contour points of the area extracted for each unit. Approximate the column by a straight line.

A region is described by an outer approximate contour and, if an island is included, an inner approximate contour corresponding to the island. It is necessary to add information to distinguish between the two, and there is a problem in that the amount of image description data (/) increases.

This invention was made to solve these problems, and when extracting regions, the extracted regions are hierarchically arranged in consideration of the adjacency and inclusion relationships between regions, and the regions are arranged in the order of the hierarchy. The present invention provides an image description method that can create image description data with a small amount of data because a region is described using only the outer approximate boundaries.

[Means for solving problems]

The image description system according to the present invention includes means for extracting a sequence of boundary points of regions constituting an image, means for detecting adjacency and inclusion relationships between regions and creating hierarchical data of the regions, and a sequence of extracted boundary points. [Operation] In this invention, adjacency and inclusion relationships between regions are detected, the extracted regions are hierarchized, and the boundaries of the regions are Approximate with primitives such as straight lines and circular arcs, and describe areas including islands using only the outer approximate boundaries.

〔Example〕.

Examples of the present invention will be described below.

FIG. 1 is an overall configuration diagram of an embodiment of this invention, (1)
is an image input means, (j) is a boundary extraction means, (3) is an area hierarchization means, (ql is a boundary approximation means, (j) is a descriptive data creation means, (6) is a descriptive data display means, (ri) is C
RT display, (t) is the original picture, (?l is the input image data, (10) is the extracted region data, (ti) is the approximate region data, and (ti) is the image description data.

t4/ In the implementation sequence shown in the figure, input image data (?) is created by converting the original 11kI image (f) into a multivalued or multicolor digital image using the image input means (1), and the boundary extraction means (co ) extracts the boundary point sequence of each region of the input image data Cd), and the area hierarchization means (,?) detects adjacency between regions and inclusion notification, and from the hierarchical data of the area and the boundary point sequence. The boundary approximation means (61) approximates the boundary point sequence of the extraction area data (10) using straight line primitives and circular arc primitives to create approximate area data (ti), and creates descriptive data. The means (j) creates image description data (/) by describing areas in the order of the hierarchy according to the hierarchical data, and the description data display means (AIK displays the contents of the image description data (12) in CR'I' This is what is displayed on the display (wa).

Figure 1 is a system configuration diagram of the embodiment, and (7) is the CR
T display, (l) is the stroke 1 detection input device, (here)
is the image memory, (3) is the processing program, (2) is the CPU, (2) is the working memory, (2) is the memory for the extraction area, and (2) is the memory for the approximate area.

(j) is a memory for one-person description.

In Figure 1, one image input device (Col) consists of an image input unit such as a television camera or image scanner, and a signal conversion unit that converts the input image signal into a signal that can be stored in the image memory (Coco). The input image data (9) is stored in the image memory (here) as a multivalued or multicolor digital image.
VC is stored. The image memories (2) are composed of RAM (random access memory) and store input image data (
8! It is used as a working area for writing flags, etc. 1. The contents of the image memory (here) are transferred to the CRT display (7) by the C'RT controller (not shown).
It is also possible to display k. The CPU 2) executes the processing described in FIG. 3 and subsequent figures in accordance with the content button of the processing program (23). Working memory (stiffness is CPU (2a)
This is a memory for recording intermediate data etc. when processing, and includes memory for extraction area (CORO) and memory for approximate area (CORO).
The image description memory (2j) is a memo that stores the extraction area data (lσ), approximate area data (ti), and image description data (here) explained in the figure E/, respectively.
It is constituted by a rewritable storage device such as AM or Winchester disk.

In addition, the resulting image description data (here) is
The data is converted into raster data by the CPU (211) or a graphics processor (not shown) and stored in the image memory (2 units), and the C! RT display (ri)
K is displayed.

FIG. 3 is a flowchart showing an outline of the process for converting input image data (9) into image description data (here) in the embodiment, and (q) is a flowchart showing the outline of the process for converting input image data (9) into image description data (here).
7) is extracted area data, (11) is approximate area data, (
1) is image description data.

In addition, Figures 1 to 7 are diagrams that supplement the explanation of Figure 3, Figure 9 is a diagram that explains region extraction processing and boundary point sequence approximation processing, and Figure 7 shows the adjacency relationship graph of regions. FIG. 6 is a diagram showing the data structure of extraction region data (10), and FIG. 7 is a diagram showing the data structure of approximate region data (11) and image description data (tS).

In step (31) in FIG. 3, a sequence of boundary points of each region constituting the input image ffl is extracted, and hierarchical data is created from K by detecting adjacency and inclusion relationships between regions. Extraction area data (10) consisting of a sequence of boundary points is created. Here, the boundary point sequence extracted in this embodiment is a boundary point sequence corresponding to the outer contour of one area, regardless of whether the area includes an island or not, in the conventional example.

For example, area A in Figure 1 (al) is defined by the outer and inner contours as a region including one island as shown in Figure 1 (g*) (bl), but in this example, area A in Figure 1 (c)
The area shown in K is defined as an area where no islands exist, that is, an area defined only by the outer contour.

Furthermore, the hierarchical data created in the embodiment is data explained below. # In the cq diagram.

To create an image of the ZILt diagram (al) using areas A', B', and C' shown in Figure 1 (cl&c), first display the large area in the 4i! Area B
', C' may be displayed in a superimposed manner. This is the fundamental principle of this invention, and the idea is to hierarchize the regions that make up the image, and in the example in Figure 9, region A is one level lower than regions B and C. Toihe.

Hierarchical data indicates the level of each area. The above will be explained in more detail with reference to Figure 3. First, the area is 1 area E-K.
This is because it is an area where two islands are integrated.

This is an area at level l below 8. Next is area E.

F and G are at the lowest level among the islands in the area, so
Defined as the area of Leperco. If we define the levels in the same way for the following regions H to K, we can see Figure 3(b).
An adjacency relationship graph for regions indicated by K is created.

The data structure of the extraction area data (lO) is shown in Figure 6 (a).
The first layer data (ri θ) consists of the hierarchical data (70) shown in Figure 6 (bl) and the boundary point sequence data (7/) shown in BL, and the first layer data (ri θ) is composed of the total number of levels of the image and the area name belonging to each level, and The boundary point sequence data (7/) is composed of the area name and the coordinate sequence or chain code sequence of each point sequence.The specific procedure of step (31) will be explained in the explanation of Figure D. .

In step (3), the extraction area data (IO) obtained by the process in step (31) is input, and
The boundary point sequence is approximated by straight line primitives and circular arc primitives, and is recorded in approximate area data (11). Through the process of step (32), each region shown in Figure 1 (C1) is approximated as shown in Figure 1 (dl).

The approximate area data (IT) is the hierarchical data (70) shown in Figure 6 (a), and the approximate boundary data (7 pieces) with the data structure shown in Figure A7 (the order in which they are stored is shown in Figure 6 (BL). Approximate boundary data consists of the area name, an operation code indicating whether it is a straight line primitive or a circular arc primitive, and coordinate data (for a straight line, the coordinate data of the start point and end point; for a circular arc, the start point, midpoint, and end point). coordinate data)
It consists of

Note that many methods have been proposed in the past for the straight-line one-circle-arc approximation method for a point sequence in step (3), similar to the method for linear approximation for a point sequence described in the conventional example (see Reference 1j). In the embodiment, any of these methods can be used. Therefore, the detailed explanation will be omitted here. @ In step (JJ), the approximate area data (11) obtained by the processing in step (3) is input and 1
Based on the hierarchical data (70), image description data (l) are created by rearranging the approximate boundary data (ri2) in order of k from the lower level busiest area. For example, in the column of the gt diagram, the areas are E, F, G, H, I.

J, is stored in acclimatization. The data structure for each area of the image description data (zz) is the same as the structure of the approximate boundary data (7 pieces) shown in the it7 diagram.

One of the methods to realize step (,?/) processing is the reference ■: IBM Journal (IBM Joa
rnaJ Re5earch and Devojop
ment Vol., 24/16A pp, adet-ri07 November (t9gco)), and only the main points of the method will be explained below.

Fig. 5 is a diagram showing the configuration of one method for realizing the process of step (31), and Fig. 5 is a flowchart showing the procedure of the method. In the figure, (?) is the input 1 data, (10) is the extraction area data, (ri/) is the rask scanning type unextracted area detection means, (lIko) is the island contour tracking means, ((I3) is the The region contour tracking means (a4A) is the region hierarchization means.

In FIG. 9, the task scanning type unextracted area detection means (I
First, in step (21), each image of the input image data (TE) is checked in the rask scanning direction. If it is determined that all pixels have been checked in step C tach), the process in step (3/) ends; otherwise, steps (zJ), (s-e)
processing is performed.

Step C! In step 3), it is checked whether an extraction flag is added to the pixel being checked. The unextracted flag is a flag that is added to pixels on the appropriate contour point sequence in the island contour tracing or area contour tracing process described later.If this flag is added and the area contains pixels, the island contour tracing process is finished. However, this indicates that area contour tracking processing has not yet been performed. If it is determined in step (j3) that the unextracted flag is not added, it is determined in step (3) whether or not that pixel is a changing point. This point of change means that the value of the pixel being checked has a gradation or color that is different from that of the area currently being scanned. Moreover, when the cough of the pixel being checked is a label, which will be described later, that pixel is not a change point. Therefore, step (33) and step (
, 1tI), if the unextracted flag is not added and it is determined that the pixel is a changing point, the pixel is included in an island that has not been tracked yet, so step (1!r) is performed.
Island contour tracking processing is performed. Otherwise, since this pixel belongs to the same area as the previous pixel, the mask scanning in step Crt) is repeated again.

In step (j), island contour tracking processing is performed by the island contour tracking means (Hiroyu). The island contour is the same as the inner contour when explaining the conventional example.

This tracking is performed using a 2x mask or a 7x, 7 mask, but during tracking, each contour point is given a label corresponding to its region name, and the contour points of the region included in the island are Adds an unextracted flag to the leftmost contour point of the scan line Kamijima. The label is written for the purpose of preventing the same island from being tracked multiple times during the determination in step (3) described above, and for the purpose of detecting adjacent regions in the region outline tracking process of step (37). Next, in step (j6), the tracing of the region outline including the pixel determined to be the change point in step (!ri) is performed in step (j6).
The tracking start point is set to that pixel as in (2).

In step (!ri), area contour tracking means C≠3
) performs area contour tracking processing. The area contour is the same as the outer contour described in the conventional example. Co×Co or JXJ masks are also used for this tracking, but during tracking, a sequence of boundary points is extracted and extracted area data (/7)K is stored, and a label of the area is written on each contour point. Add extraction flag.

Detect labels for regions that touch that region. Pixels to which an unextracted flag is added in this process are pixels in the contour point sequence of the area that is in contact with the area and for which contour tracking has not yet been performed (thus, the pixel value corresponds directly to the gradation or color). So, the pixel currently being tracked and the mask direction 111! l (facing the right side)
This is a pixel that is in contact with K.

In step (3g), an adjacency relationship graph as shown in the ysS diagram (blk) is created sequentially from the labels of the adjacent regions detected in step (!ri), and hierarchical data (i.e., 0) is generated. When step (!t) is completed, the mask scanning of step (51) is (repeated).

The method of the descriptive data display means (6) for displaying the conventionally created image descriptive data (12) is as follows.

For example, there is a conventional example (Reference [4!]: 1985 IEICE General Conference Proceedings ``A Proposal of a Multicolor Single Graphic Command Coding System''), and its explanation will be omitted here.

According to the embodiments of the present invention as described above.

Input image data (9) in the process of step (31)
As well as extracting the boundary point sequence of the region that makes up the area.

Hierarchical data by detecting adjacency and inclusion relationships between areas (70)
Create a step (3) in the process of step (3).
)) is approximated by straight line primitives and circular arc primitives, and in the process of step (3J), the hierarchical data (70) obtained in the process of step (Ji) is approximated. This has the advantage that image description data (12) with a smaller amount of data can be created, rather than describing regions in the order of hierarchy using outer approximate boundaries.

Further, according to the embodiment of the present invention, in approximating the boundary of the area where the island exists, approximation to the inner boundary is not necessary, and the approximation processing time is shortened.

Further, according to an embodiment of the present invention, the descriptive data display means (
In 6), it is sufficient to display only the closed region in which no islands exist, and the procedure for displaying one is simplified.

In the above embodiment, a CRT display was used as the display device, but a liquid crystal display, a plotter, etc. may also be used. Furthermore, in order to make the explanation easier in the system configuration shown in Figure 1, the working memory (Kori), the extraction region memory (Koro), the approximate region memory (27), and the image description memory (Kot) are separated. Although shown in the figure, it may be configured in a physical memory device, and in each process unnecessary information may be erased and new information may be created thereon.

Further, in the process of step (32), a straight line or a circular arc is used as the approximated primitive, but the image may also be described using other primitives such as a clear circle or a spline curve.

Also, in the process of step (3/), the result of the pixels located at the outermost position of the area (outer contour point sequence) was used as the boundary point sequence to be extracted, but in @10(b), the thick '% line and Other boundary point sequences, such as a sequence of points on a digital grid located between each contour point indicated by a 0" mark, may also be used.

Furthermore, in the above embodiment, the areas are rearranged in the order of hierarchy based on the hierarchy data (RIO) in the process of step (33);
This process is omitted and the approximate area data (/l) is used as the surface description data (l), and the descriptive data display means (6)
The configuration may be such that the areas are displayed in the order of hierarchical data (RIO).

Furthermore, although the above embodiment has been described as a system that displays image description data (12), image description data (12)
can be used for systems that perform more advanced processing such as pattern recognition and image analysis, and it not only produces the same effect as the above implementation sequence, but also combines this single wound description data (1) and hierarchical data (R). Q) There is an effect that information regarding the adjacent one-layer structure of images can be obtained from the button.

〔Effect of the invention〕

As explained above, the present invention extracts a sequence of boundary points of regions constituting input th image data, detects adjacency and inclusion relationships between the areas, creates hierarchical data, and extracts a sequence of boundary points from the extracted boundary points. By approximating with primitives such as straight lines and circular arcs, and describing the area in the order of the first layer data button and the outer approximate boundaries, image description data with a small amount of data can be obtained.

[Brief explanation of the drawing]

FIG. 1 is an overall configuration diagram of an embodiment of the present invention, FIG. FIG. 9 is a diagram explaining the region extraction process and the boundary point sequence approximation process. FIG. 3 is a diagram explaining the region adjacency relationship graph. FIG. 6 is the data of the extracted region data, #R Figures showing the composition and g diagrams are approximate area data and +
Figure 81 is a diagram showing the data structure of m-image description data; Figure 81 is a block diagram of one method for realizing the process of step (31) in Figure 3;
Figure 1 is a flowchart showing the steps of Takahiro's method.
Fig. 210 is a diagram showing another boundary point sequence extracted in the embodiment of the present invention, Fig. taii is a system configuration diagram of the conventional example, Fig. 1J is a flowchart showing the procedure of the conventional example, and Figs. 1 and 7 are FIG. 3 is a diagram illustrating the outline of a region. In the diagram, (/l is an image input means, (c) is a boundary extraction means, (J) is a region hierarchization means, (ri is a boundary approximation means. (j) is a descriptive data creation child actor, (1) is Descriptive data display means, (ri) is a CRT display, (g) is an original image,
(te) is input 1 data, (to3 is extraction area data,
(11) is approximate area data, (i) is +ili image description data, (l) is image input device, (here) is IIfI
i-ku memory. (3) is the processing program, (3) is the CPU, (3)
) is working memory, (coro) is memory for extraction area,
(Koku) is the approximate region memo I), (2g> is the memory for image description, (,7/) to (33) are each step in the flowchart diagram in Figure 3, and (70) is hierarchical data. (Ri/ ) is boundary point sequence data, (Kuko) is approximate boundary data, r Hiro/) is Tesk scanning type unextracted area detection means, (Kuko) is island contour tracking means, (173) is area contour tracking means, (Hiroda ) is the area hierarchization means, (31) to (3g) are each step in the flowchart of Fig.
) to (ko/J) are each step in the flowchart diagram of @/2, (2ir) is the co-valued image data, (ko/4
) is contour point sequence data. In each figure, the same reference numerals indicate the same or corresponding parts. Figure 2 Figure 4 (a) (C) Figure 4 (d) Figure 5 (b) Figure 10 (b) W) 13 procedural amendments (voluntary) Showa 6゜na. January 4th

Claims (1)

[Claims] In an image description device that describes the boundaries of connected regions constituting an image using primitives such as straight lines and circular arcs, means for extracting a sequence of boundary points of the connected regions, and adjacency and inclusion between the connected regions. An image description device comprising: means for detecting a relationship and creating hierarchical data of the connected region; and means for approximating the boundary point sequence with a primitive such as a straight line or an arc.