JPS6084678A - image recognition device - Google Patents

Abstract

PURPOSE:To speed up the structure analysis and feature extraction of binary image data by labeling a closed area having a polarity different from a closed area composed of a recognized body as well as the recognized body. CONSTITUTION:A data string part S overlaps with a data string part (n), but the data string part (n) already overlaps with a data part (q) to cause divergence. For the purpose, a column number 5 and edge data (s) are made to correspond to an edge number 8, and the column number 2 of the data string part (n) is stored in the part DF1 of the column number 5 correspondingly. Further, the column number 5 of the data string part (s) is made to correspond to the part DF2 of the column number 4 of a data string part (r). When the value of the CF1 or DF1 of a column (i) is (j), columns (i) and (j) belong to the same closed area and when both values of the CF2 and DF2 of the column (i) are both (k), the constituted area I is included in the closed area (k) to which the column (k) belongs.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a general-purpose image recognition device for performing structural analysis and feature extraction of binary digital images.

Configuration of conventional example and its problems Conventionally, a binary digital data sequence with polarity of "0" and "1" obtained by horizontally scanning the recognition target image is input, and each closed position in the recognition target image is input. When labeling is performed by assigning the same number to the data string part, which is a continuous data string of the same polarity on each horizontal scanning line belonging to an area, only the closed area formed by the recognition object in the recognition target image is labeled, Labeling is not performed on closed regions that have a polarity different from the closed region formed by the recognition object, such as the background of the recognition object or a hole in the recognition target image. In addition, since it was not possible to determine the inclusion relationship between closed regions of the above polarities in a single horizontal scan, it was necessary to calculate the feature amount of the closed region included in each closed region or the closed region included. It was necessary to scan the screen several times, which was very time consuming.

FIG. 1 shows an example in which a recognition target image (camera field of view) is scanned and a closed area (shaded area) that is a recognition object is labeled. FIG. 2 shows the labeling results. In this conventional method, even if the data string portion of each scanning line as shown in FIG. 2 is used, it is not possible to directly obtain the number, area, boundary length, etc. of closed regions included in the closed region.

OBJECTS OF THE INVENTION It is an object of the present invention to solve the above-mentioned problems and to provide an image recognition device that can determine the inclusive relationship between closed regions in one scan.

In order to efficiently perform the labeling, structural analysis, and feature extraction of the constituent images of the invention, it is necessary to simultaneously perform the structural analysis and feature extraction while labeling the data string of each scanning line.

The image recognition device of the present invention not only labels a closed region composed of a recognition object in a recognition target image, but also labels a closed region composed of a recognition object, such as a background or a hole of the recognition object, with a polarity different from that of the closed region composed of the recognition object. At the same time, the closed regions that the system has are labeled, the inclusion relationships between each closed region are determined in real time, and the characteristic quantities of each closed region are simultaneously determined, allowing for structural analysis and feature extraction of binary image data at extremely high speed. This information is used to perform image recognition at high speed.

DESCRIPTION OF EMBODIMENTS FIG. 3 is a block diagram showing the configuration of an image recognition apparatus according to an embodiment of the present invention. In the figure, 10 is a signal input terminal, and an image scanning signal obtained by scanning a recognition target image with an image scanning device is input to this input terminal 10. Reference numeral 11 denotes a binarization unit that converts this image scanning signal into binary digital data (hereinafter simply referred to as data example).

Reference numeral 100 denotes a labeling structure analysis unit that analyzes in real time the structure of an image representing the inclusion relationship of closed regions in the binary recognition target image. 12 is the point where the data changes in the data string, i.e. "1" from rOJ or "0" from rIJ.
This is an edge detection section that detects an edge point that changes to ``.

This edge detection unit 12 detects points a, b, and c as edge points when the data string of the i-th scanning line is input, for example, in recognition of a recognition target image as shown in FIG.
An edge detection signal 16 is output. These edge points can represent continuous data strings of the same polarity on the horizontal scanning line. 1
3.16 is a one horizontal scanning line delay unit that delays the scanned data string by one horizontal scanning line; 14 compares the polarity of the data on the current horizontal scanning line with the data one horizontal scanning line before; If the two data are of the same polarity and are both "1" (or "0"), the overlap detection section outputs an overlap signal 17. This overlap detection section 14 is configured to detect, for example, the fourth
In the figure, point d (=404
) and the point c(-[oJ) on the i-th scanning line are data of the same polarity, so an overlap is detected and an overlap signal 17 is output.

Reference numeral 18 denotes a closed area top detection unit which outputs a top signal if the continuous data string portion of the same polarity of the horizontal scanning line is the top of the closed area.

For example, in FIG. 5, this closed area top detection unit 18
Since the data string portion of the first scanning line is the data string portion on the first horizontal scanning line of closed area 1, it is detected as the top of closed area 1. After scanning the first scanning line, the top detection signal 21 is output. Furthermore, the data string portion of the closed area 2 on the first scanning line is
-1) It does not overlap with any data string portion on the scanning line and becomes the top of the closed area 2, and after scanning the data string portion of the closed area 2 on the first scanning line, outputs the top detection signal 21. 19, the data string portion on the current horizontal scanning line overlaps with the data string portion one horizontal scanning line before, but the data string portion one water scanning line before is already overlapped with another data string portion of the current horizontal scanning line. Detects the overlapping divergence state and outputs the diver fence detection signal 2.
This is a divergence detection section that outputs 2. In other words, the divergence detection unit 19 is configured such that, for example, in FIG. detects that a divergence occurs when it overlaps with data string part B before it overlaps with data string part B,
A divergence detection signal 22 is output.

20, the data string part on the current horizontal scanning line overlaps with the data string part one horizontal scanning line before, but before the data string part on the current horizontal scanning line overlaps with the data string part one scanning line before,
A convergence state that overlaps with another data string section in front of the scanning line is detected, and convergence detection (i No. 23
This is the output convergence detection section. That is,
For example, in FIG. 7, the data string section H on the i-th scanning line overlaps the data string section G on the (i-1)th scanning line, but the convergence detection section 2o It is detected that convergence has occurred when H overlaps with the data string portion E on the (l-1)th scanning line before it overlaps with the data string portion G on the (i-1)th scanning line, and a convergence detection signal is generated. Outputs 23.

24 is an edge counter that counts the number of the edge, that is, the number of the data string, when the edge detection signal 16 is input, and also counts the horizontal address of the edge and the number of the current horizontal scanning line. Department.

26, when the closed area top detection signal 21 is input in the data string part on the current horizontal scanning line, increases the column number given to each data string part by 1, and gives the above column number to the data string part. In addition, when the divergence detection signal 22 is input in the data string section on the current horizontal scanning line, the column number is incremented by 1 and becomes the column number of the data string section, and a closed area is added to the data string section on the current horizontal scanning line. If neither the top signal nor the divergence detection signal is input, a column number with the same value as the column number of the data string part one horizontal scanning line before the data string part overlapped with the data string part is given to the data string part, and This is a column number setting unit that assigns the start of the column, the end of the column, and the polarities of rlJ and roJ of the data string to the data string.

In other words, the column number setting section 25 sets the data string portion B on the first scanning line to the (i-1)th column in FIG. 6, for example.
When the data string section A overlaps with the data string section A on the scanning line, and neither the closed region top detection signal 21 nor the divergence detection signal 22 is output, the column number of the data string section A is changed to the data string section B.
granted to.

Furthermore, since the data string portion C on the i-th scanning line is the top of the closed area, the column number is incremented by 1 and assigned to the data string portion C, and a column start flag is also added. Further, a divergence is detected in the data column portion on the i-th scanning line, and the column number is incremented by 1 and assigned to the data column portion, and a column head flag is assigned.

Furthermore, corresponding polarities are assigned to the data string portions A, B, Q, ]).

26 is a flag DF indicating the concatenation state of each data string when the divergence detection signal 22 is input in the data string portion on the current horizontal scanning line. The column number of the data string section is stored in correspondence with the column number of the data string section, and the column number of the data string section is stored as a flag DF2 indicating the inclusion relationship of each data string section. This is a divergence flag setting section that stores the divergence flag in correspondence with the following. That is, this dyno (-sience flag setting section 26 is set to
Dyno (-) is occurring in the data string portion of the scanning line, but the column number of the data string portion A is stored in the flag DF in correspondence with the column number of the data string portion, and the column number of the data string portion A is The column number of is made to correspond to the column number of the data string portion C and is stored in the flag DF2.

When the convergence detection signal 23 is input in the data string section on the current horizontal scanning line, 27 is used as a flag CF indicating the concatenation state of each data string section, and also indicates the column number of the data string section. If an overlap is newly detected, it is stored in correspondence with the column number of the data string part one scanning line before, and the column number of the data string part one scanning line before is stored as a flag CF2 indicating the inclusive relationship of each data string part. This is a convergence flag setting unit that stores a number in correspondence with a column number of a data string section immediately before the data string section one scanning line before.

For example, in FIG. 7, convergence has occurred in the data string section H of the i-th scanning line. The column number is memorized, and the column number of the data string part G is stored in the data string part F.
is stored in the flag CF2 in correspondence with the column number of.

28 inputs the output signal of the binarization section 11, and according to the column number of the data string section on the current scanning line determined by the column number setting section 26, the area, moment,
This is a feature extraction unit that calculates the boundary length.

29 is a column number setting section 25, a divergence flag setting section 26, a convergence 7 rasog setting section 27,
This is a memory section that stores data obtained by the feature extraction section 28, the edge counter section 24, and the edge detection section 12.

For example, as shown in FIG. 8, the memory unit 29 stores, in correspondence with column numbers, respective closed area numbers OF, level numbers LF indicating the level of inclusion relationship, master numbers MS indicating included closed area numbers, and inclusion numbers. Slave number SL indicating the closed area number to be used, convergence flag OF, , OF2
, divergence flag DF, , DF2 and area, first order.

As shown in FIG. 9, the edge data representing the beginning and end 7 rasogs of the column, the column number and the data string part are stored in correspondence with the edge numbers. As shown in FIG. 10, edge numbers are stored in correspondence with scanning line numbers.

Reference numeral 30 denotes a recognition unit that organizes information requested when processing all pixels and recognizes the image, and is constituted by a microcomputer.

The recognition unit 30 recognizes the CF+ + D Fl, CF2
, D F2 is used to obtain the closed region number of each column, the SL pointing to the number of the closed region included in the closed region, and the MS pointing to the number of the closed region included in the closed region. , MS and SL, the inclusive relationship between each closed area is easily clarified, and a level flag LF representing the inclusive relationship between each closed area in FIG. 8 is set.

Furthermore, the recognition unit uses the features calculated for each column,
The sum of the column feature amounts is calculated for each closed region number, and the feature amount for each closed region is calculated.

Regarding the apparatus of this embodiment configured as above, the following will be explained.
Let's explain its operation.

An example of a recognized image shown in FIG. 11 will be explained.

Figure 11 is a diagram showing the image data after the image signal has been binarized.The shaded area is the recognized object, which has a value of "1," and the blank area is the background, which has the rOJO value. Assume that it has one hole. It is also assumed that the data string portions in blank areas and shaded areas are represented by respective edge points.

Now, in FIG. 11, the data string portion 4 of the first scanning line is a blank space and is the first data string portion, so as shown in FIG. 13, the column number is 1, the edge data is e, and the polarity is 7. O for Ranog, 1 for the first flag of the column.
is stored in correspondence with edge number 1. Also, since the edge number of the first scanning line is 1, the scanning line number 1 in FIG.
is associated with 1 and stored.

The data string portion m of the second scanning line overlaps with the data string portion m. Therefore, the same column number as data string part 4 is stored in correspondence with edge number 2 in FIG. 13. Also, since the edge number of the second scanning line starts from 2, 2 is associated with the scanning line number 2 in FIG. 14 and stored. Next, the data string part n is the shaded part and is the top of the closed area. Therefore, the column number is increased by 1, n is the edge data, and 1 is the polarity flag. 1 is stored in the head flag of each column in correspondence with edge number 3 in FIG. 13.

Next, data string portion 0 overlaps with data string portion e, but data string portion l already overlaps with data string portion m, resulting in a divergence. Therefore, increase the column number by 1 and set it to 3.
It is stored as shown in FIG. 13, and the DF of column number 3 in FIG. 12 corresponds to column number 1 of the data string section l.
The DF2 section of column number 2 of data string section n corresponds to column number 3 of data string section 00.

Next, the data string part p of the third scanning line overlaps the data string part m. In FIG. 13, the edge number 6 corresponds to 1 as the column number and p as the edge data length. 6 is stored in correspondence with the scanning line number in FIG. Also, the data string part q overlaps with the data string part n, and in FIG.
, q are stored in correspondence as edge data. On the other hand, the data string portion r does not overlap with any data string. Therefore, the column number is increased by 1 to 4, the head flag of the column is set, and the flag is stored in correspondence with edge number 7 in FIG.

Further, the data string portion S overlaps with the data string portion n, but the data string portion n already overlaps with the data string portion q, causing a divergence. Therefore, edge number 8 in FIG. 13 is associated with column number 6 and edge data with S, and furthermore, in FIG. 12, DF of column number 5,
The column number 2 of the data string section n is stored in association with the column number 2 of the data string section n. Further, the DF2 section of column number 4 of data string section r is made to correspond to column number 6 of data string section S, and data string section t overlaps with data string section 0 and is stored in correspondence with edge number 9 in FIG. .

Next, the data string portion U of the fourth scanning line overlaps the data string portion p, and the column number 1 corresponds to the edge number 10 in FIG.
The scanning line number 4 in the figure is made to correspond to 10, the data string portion V overlaps the data string portion q, and the edge number 11 is stored in correspondence to 2 as a column number. Furthermore, the data string portion V also overlaps the data string portion S, resulting in convergence. Therefore, in FIG. 12, column number 2 of data string section V corresponds to CF of column number 5 of data string section S, and column number 6 of data string section S corresponds to CF2 of column number 4 of data string section r. It is stored in correspondence with the section. Also, in Figure 13,
The column to which the data string portions r, s belong ends, and a column end flag for the data string portions r, s is added. Furthermore, the data string portion W overlaps with the data string portion t. Therefore, the number 3 is associated with the edge number 12 in FIG. 13 and stored.

Next, the data column section X of the fifth scanning line overlaps with the data column section U. Therefore, 1 is associated with the edge number 13 in FIG. 13 as a column number and stored. Furthermore, the data string section X has an overlapping convergence with the data string section w. Therefore, in FIG. 12, column number 1 of the data string section
2 and are stored in correspondence with each other.

Furthermore, since the data string sections v, w, and x end, respective column end flags are added in FIG. 13.

A flag OF representing the connected state in which more than one cent has been sent;
, DF, ,Flag representing inclusion relationship OF2゜DF2
By using the OF, LF, MS.

Each flag of SL can be set.

In other words, if the value of OF or DF of column l is j, then column i and column j belong to the same closed region, and if the value of both OF2 and DF2 of column i is k, then column i and column j belong to the same closed region. The closed area K that is constructed is included in the closed area K that belongs to the column, and is included in the closed area K that belongs to the column.

The LF, MS, and SL flags are displayed as shown in Figure 16. In other words, according to FIG. 15, columns 1 and 3 and columns 2 and 6 are the same closed area, and the closed area 2
is included in closed area 1, closed area 3 is included in closed area 2, and closed area 2, that is, the recognition object is included in one hole area 3.
It can be seen that the structure of FIG. 11 is the same as that of FIG. 16.

Furthermore, in the above processing, the area of each data string part,
Prepare the first moment, second moment, and boundary length, and put them into the area of the same column as the edge data column number. After all data is completed, add the feature quantity with the same closed region number to calculate the area for each closed region. A feature quantity can be divided.

Effects of the Invention With conventional labeling methods, it was necessary to scan image data several times in order to analyze the structure of the target object, but with the present invention, the inclusion relationship of closed regions in the target image can be determined by scanning the image once. In addition, it is possible to easily and quickly determine the characteristic amount of the closed area that each closed area includes or is included in.

In addition, by associating column numbers with edge data, and by associating column numbers with closed region numbers, you can see which closed region each edge data belongs to, which is very effective for calculating more detailed feature quantities. It is.

[Brief explanation of drawings]

Figure 1 is an explanatory diagram showing an example of a conventional labeling method;
The figure is an array diagram showing an example of data obtained by the labeling method shown in Fig. 1, Fig. 3 is a block diagram showing an embodiment of the image recognition device in the present invention, and Fig. 4 is an overlap detection in the same embodiment. FIG. 5 is an explanatory diagram of the closed region top detection section, FIG. 6 is an explanatory diagram of the divergence detection section,
FIG. 7 is an explanatory diagram of the convergence detection section, FIGS. 8 to 10 are schematic diagrams for explaining the memory section, and FIG. 11 is an array diagram showing an example of image data input to the device of the embodiment, Figures 12 to 14 are schematic diagrams showing the contents of the memory section, Figure 16 is an array diagram showing feature data after reassigning closed area numbers, and Figure 16 is a diagram showing the hierarchical structure of image data. It is. 10...Input terminal, 11...-Binarization unit, 29...Memory unit, 30...Recognition unit, 100...Structure Analysis department. Name of agent: Patent attorney Toshio Nakao and 1 other person No. 1
Figure 4 Figure 6 Figure 8 Figure 9 Figure 10 Figure 12 Figure 131 Me1

Claims (1)

[Claims] A binary digital text string obtained by horizontally scanning an image to be recognized that includes at least one closed region is input, and a continuous data string portion of the same polarity of the current horizontal scanning line is used as an input. Check for overlap with the data string part of the previous horizontal scanning line, and if there is no overlap, increase the column number representing the set of data string parts by 1 and add it to the data string part of the horizontal scanning line. and the column number of the data string part one horizontal scanning line before, when at least a part of the continuous data string part of the current horizontal scanning line of the same polarity overlaps with the data string part one horizontal scanning line before. a second means for assigning the same column number to the data string part;
The consecutive data strings of the same polarity of the current horizontal scanning line overlap with the data string of the previous horizontal scanning line, but the data string of the previous horizontal scanning line has already overlapped with the other data string of the current horizontal scanning line. In the case of a divergence that overlaps with , the column number is incremented by 1 to be the column number of the data string section, and the flag DF indicating the connection state of the data string section and the data string section one horizontal scanning line before is set as the flag DF. The column number of the data string section before the horizontal scanning line is stored in correspondence with the column number of the data string section, and furthermore, a 7 rasogogram representing the inclusive relationship between the data string section and the data string section immediately before the concerned data string section is stored. A third DF2 stores the column number of the data string section in correspondence with the column number of the data string section immediately before the data string section.
, the continuous data string portion of the current horizontal scanning line of the same polarity overlaps with the data string portion of the previous horizontal scanning line, but the data string portion overlaps with the data string portion of the previous horizontal scanning line. In the case of convergence, which has already overlapped with another data string part one horizontal scanning line before the overlap, the overlap with the current data string part is newly detected with the data string part one horizontal scanning line previous. Flag CF indicating the state of connection with the current data string section
, the column number of the current data string section is stored in correspondence with the column number of the one horizontal scanning line before, and the data string section of the one horizontal scanning line before and immediately before the data string section of the one horizontal scanning line before is stored. Flag C indicating the inclusion relationship with the data string part of
a fourth means for storing, as F2, a column number of the data string section one horizontal scanning line before in correspondence with a column number of a data string section immediately before the data string section one horizontal scanning line before; A fifth means of adding the area, boundary length, first moment, and second moment of each data string part to the feature item area corresponding to the force number of the data string part of the line, and adding all the horizontal scanning lines. After processing the data string by the first to fourth means, each of the flags OF. DF,,CF2. Based on DF2, the number of closed regions of the target image, the inclusion relationship of each closed region, and the structure of the image. A sixth method that calculates the depth of the structure and calculates the features of each closed region by adding the features calculated for each column by the fifth means to each column having the same closed region number. An image recognition device characterized by comprising: means.