JPH0628472A - Method and device for recognizing image - Google Patents

Abstract

PURPOSE:To improve the recognition rate even in the case where a pattern is deformed or missing by generating a network from a relative position relation of a rule of the pattern, and deciding from a structure of this network. CONSTITUTION:From an image signal having patterns arrayed by a prescribed rule, outlines of the patterns are extracted, respectively, a centroid position of a range surrounded by the outline is detected, the centroid position becomes a node, a segment for connecting an arbitrary centroid position and the centroid position being adjacent thereto become an arc, and from a prescribed network setting start point, a network is enlarged successively based on a relative position relation condition determined in advance based on the rule, and based on this network, an array of an image is recognized.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image for recognizing an array having a known regularity and a certain regularity such as the recognition of the stitch arrangement of the knitted fabric, the defect detection of the knitted fabric, the detection of the crystal structure and the like. The present invention relates to a recognition method and an image recognition device thereof.

[0002]

2. Description of the Related Art As a method for recognizing a pattern arrangement,
The position of the center of gravity of the pattern is detected from each of the image signals, the position of the center of gravity is used as a node, and the line segment connecting an arbitrary position of the center of gravity and an adjacent position of the center of gravity is used as an arc to set a network of nodes and arcs. There is known a method of recognizing a stitch array image by utilizing the relaxation method (Japanese Patent Laid-Open No. 2-310781).

[0003]

In the above method, since the image is recognized based on the set network, it is impossible to recover the error recognition at the stage when the object is binarized and extracted or the adjacent relation is extracted. There was a problem. For example, when the position of the center of gravity is erased due to the overlapping of the knitted fabric or the fray of the yarn, or the portion which should originally be one position of the center of gravity is divided into two, there is a problem that the recognition rate is lowered. Further, although the stitches of the knitted fabric knitted with the plain cloth are regularly arranged, it is difficult to accurately recognize the arrangement of the stitches because the knitted fabric itself is easily deformed.

Therefore, an object of the present invention is to provide a more reliable recognition method and its apparatus.

[0005]

In order to solve the above-mentioned problems, the first aspect of the present invention is to extract the contours of the pattern from an image signal having a plurality of patterns arranged according to a certain rule. Detect the respective barycentric positions in the area enclosed by the contour, use each barycentric position as a node, and use the line segment that connects any barycentric position and its adjacent barycentric position as an arc. Based on a predetermined relative positional relationship condition, the network is sequentially expanded, and the arrangement of images is recognized based on the network.

According to a second aspect of the present invention, the image inputting means, the contour line extracting means for extracting the contour of the pattern from the image signal input from the image inputting means, and the center of gravity of the range surrounded by the contour. A barycentric position detecting means for detecting and storing a position, a barycentric position as a node, and a line segment connecting an arbitrary barycentric position and a barycentric position adjacent thereto as an arc, a predetermined for setting a network of nodes and arcs Network setting start point determining means for determining the node of, and the network is sequentially expanded from the network setting start point based on a predetermined relative positional relationship condition,
A network setting unit that sets a network within the range of the image signal, a recognition unit that recognizes an array of images based on the set network, and an output unit that outputs a result recognized by the recognition unit. An image recognition device characterized by the following.

[0007]

According to the present invention, the most regular part of the image with less unclearness and noise is set as the network setting start point, and the relative positional relationship condition determined in advance from this network setting start point based on a certain rule. Network is expanded sequentially to form a network, so even if there are unclear parts or noise in the image, it is possible to prevent the network from being formed by mistake and form a network that is faithful to the regularity and accurate. be able to. Therefore, accurate recognition can be performed based on this network.

[0008]

The present invention will be described in detail below with reference to the drawings. FIG. 1 is a diagram showing the basic configuration of the present invention. In FIG. 1, reference numeral 100 is an image input means for inputting an image signal of a pattern to be recognized.

Reference numeral 200 is a contour line extraction means for binarizing the image signal input from the image input means to extract the contour of the pattern. Reference numeral 300 denotes a center-of-gravity position detection unit that detects and stores the center-of-gravity position in the range surrounded by the contour lines of the pattern extracted by the contour line extraction unit. 400 is a network setting start point for determining a predetermined node for setting a network of nodes and arcs, with the center of gravity position as a node, and a line segment connecting an arbitrary center of gravity position and an adjacent center of gravity position as an arc It is a means.

Reference numeral 500 denotes a network setting for confirming the network based on a predetermined relative positional relationship condition from the starting point of the network setting or expanding the network sequentially while correcting the network and setting the network within the range of the image signal. It is a means. Reference numeral 600 is a recognition means for recognizing the array of images based on the network set by the network determination means, and calculates, for example, the angle of the arc direction.

Reference numeral 700 denotes an output means for outputting the result recognized by the recognizing means, which displays the arrangement direction of the pattern as an image and stores the recognized result. FIG. 2 is a diagram showing a specific circuit configuration of the embodiment of the present invention.
In FIG. 2, a central processing unit (CPU) 1, an image processing processor 2, an image processing memory 3, a read only memory (ROM) 4, a random access memory (RA).
M) 5, an input / output interface (I / O) 6 is connected to the bus 7, and the I / O 6 has an imaging device 8, a keyboard 9, and a display device 1.
0 is connected.

The CPU 1 controls the entire apparatus and gives operation instructions to the above-mentioned respective parts. The image processor 2 as the contour line extracting means, the center-of-gravity position detecting means, the network setting start point determining means, and the network setting means performs image processing on the image obtained from the image pickup device 8 to obtain the contour line extracting processing and the center-of-gravity position. Performs detection processing and network creation processing.

The network setting means and the network angle calculating means can be processed by the CPU 1. An image signal from the image pickup device 8 is stored in the image processing memory 3. The ROM 4 pre-stores control programs executed by the CPU 1 and the image processor 2.

The RAM 5 temporarily stores the calculation data of the CPU 1 and the image processor 2. I / O 6 is connected device and C
Information is transferred to and from PU1. The image pickup device 8 as an image input unit is for picking up an image of an object by, for example, a solid-state image pickup element (CCD) and converting the target object into an image signal by photoelectric conversion. The image signal may be input from a storage medium such as a CD-ROM.

An operation instruction for the CPU 1 is input from the keyboard 9. The display device 10 displays the picked-up object image and the image-recognized object array. FIG. 3 is a flow chart showing the operation procedure of the embodiment of the present invention.
FIG. 3 is an example of recognition of the stitch arrangement of a knitted fabric knitted with a plain cloth. The stitch array of the knitted fabric knitted with the plain cloth has a hexagonal structure, and an image having a plurality of patterns arranged in a certain rule in this manner is recognized. In this embodiment, the directionality of the array is recognized.

The operation of this embodiment will be described below with reference to the flow chart of FIG. (1) Pattern contour extraction First, the image signal of the stitches output from the imaging device 8 is I / O.
It is stored in the image processing memory 3 via 6. The image signal is obtained by receiving the illumination light from the back of the knitted fabric 11 through the knitted fabric as shown in FIG. 17, and when the knitted fabric is imaged, the yarn portion is black and the stitch portion is white. . This white portion has a regular hexagonal structure as shown in the stitch pattern diagram of FIG.

Then, various tables such as the node center of gravity table (FIG. 4), the adjacent point table (FIG. 5) and the arc table (FIG. 6) are initialized (step S10).
0). Note that each table will be described in detail later. Next, thinning is performed to extract the contour from the image signal.
In the thinning process, for example, an image signal, which is a density signal, is binarized by comparing it with a prescribed value, and a contour line can be extracted by thinning the black part, that is, the knitting yarn part.
This contour line information is stored in the RAM 5 (step S1).
20).

(2) Detection of barycentric position The image processor 2 detects the barycentric position of the closed section (stitch) surrounded by the contour line based on the contour line information stored in the RAM 5. For example, the total of coordinates and the area of the closed section surrounded by the contour line can be obtained, and the position of the center of gravity can be obtained from them. In this way, each time the center of gravity position of the stitch is detected, a number (node number) is given to the detected center of gravity position (node) for network creation, and in the form of the node center of gravity table shown in FIG. The node number and its barycentric coordinates (x, y) are stored in the RAM 5 (step S
130).

Next, for each node, the adjacent nodes around it are obtained, and in the form of the adjacent point table shown in FIG.
Each node number and the node numbers adjacent to it are stored in the RAM 5. (Step S140). (3) Selection of network setting start point In order to create an accurate network, the most regular part with less unclear image and noise is used as the network setting start point (start point of network creation). In this embodiment, the center node having the most regular hexagonal structure is selected using the node center of gravity table and the adjacent point table, and is used as the network setting start point. As the network setting start point, the node of the most regular part may be selected while looking at the image, but in the present embodiment, the start point is determined as follows.

First, an arbitrary node K is selected, and the nodes adjacent to the node K are sorted in the adjacency order (clockwise). Then, as a result of the sorting, when there are 6 nodes, the selected node K is extracted. Similarly, for the adjacent nodes located around the extracted node K, the nodes adjacent to the adjacent node are sorted, and it is checked whether or not the adjacent node has 6 adjacent nodes. Remember as the level of regularity. Such a process is performed for all the nodes, and the node with the highest regularity, that is, the node with the largest k is used as the starting point for network creation. If there are a plurality of nodes with the same k, the node near the center of the image is selected (step S150).

(4) Network Creation When the start point of network creation is determined in step S150, a network of nodes and arcs is created using the node center of gravity table and the adjacent point table. Network growth is performed when the most certain condition (regularity is consistent) (relative positional relationship condition) is met according to a previously known regularity, and then gradually (including regularity is inconsistent) condition ( The network is grown under the relative positional relationship condition), and these processes are repeated and completed until the network does not grow for all nodes and arcs. The relative positional relationship condition is based on the positional relationship between nodes and the positional relationship between arcs and nodes.

The creation of the network will be described below by showing detailed procedures and conditions. (4-1) Direction Labeling As shown in FIG. 7, the three directions a, b, and c of the three sides of the triangle from the start point of network creation to adjacent nodes
To decide. For example, in FIG. 7, the upper right-lower left direction of the drawing is the a direction, the upper-lower direction of the drawing is the b direction, and the upper left-lower right direction of the drawing is the c direction (step S160).

Next, a label (A, B, C) indicating a direction is given to a line segment (arc) with an adjacent node located in the direction determined in step S160, and each node is centered for each direction. RAM5 sequentially in the form of arc table
To remember. This direction label is given every time the network grows. As shown in FIG. 6, the arc table is divided into three tables for each direction, and each table is stored in the form of a node number and a node number adjacent to that node. For example, it can be seen that the arc of label A exists between the node 1 and the node 4 and between the node 5 and the arc of the label B exists between the node 21 and the node 3. Further, it can be seen that the arc of label A exists between the node 2 and the node 3 and the arc of label B exists between the node 2 and the node 20.

(4-2) Growth of Network (Condition 1) The following conditions 1-1 to 1-3 in the triangle around the known part.
The part that conforms to is added to the arc table as a correct network and stored (step S170).

[0025]

[Table 1]

When the network is grown under these conditions 1-1 to 1-3 and there are no more nodes or arcs to process, the process proceeds to the next condition 2. (4-3) Network contradiction detection (condition 2) The following conditions 2-1 are added to the network grown under the above conditions.
Determine if there is any contradiction. The inconsistent part is deleted from the arc table (step S180), and the following condition 2 is satisfied.
The arc is labeled with -2 and used as new estimation information (step S190).

[0027]

[Table 2]

When the processing is performed under the condition 2 as described above, the condition 1 is returned to again to grow the network. When it becomes impossible to grow a new network by performing the processing under the conditions 1 and 2, the process proceeds to the next condition 3. (4-4) Network Correction (Condition 3) Regarding the network grown under the above Conditions 1 and 2, the network is corrected according to the following Condition 3 (step S200).

[0029]

[Table 3]

When the network is grown under the condition 3 and there are no nodes or arcs to be processed, the condition 1 is re-established.
Return to and grow the network. When it becomes impossible to grow a new network by performing the processing under the conditions 1 to 3, the process proceeds to the next condition 4. (4-5) Growth of Network (Condition 4) Regarding the networks grown under the above-mentioned Conditions 1 to 4, a portion conforming to the following Condition 4 is added to the arc table as a correct network and stored (step S).
210).

[0031]

[Table 4]

When the network is grown under this condition 4 and there are no more nodes or arcs to be processed, the condition 1 is re-established.
Return to and grow the network. When it becomes impossible to grow a new network by performing the processing under the conditions 1 to 4, it is assumed that the network has been formed for the nodes and arcs on the screen, the network is determined, and the process proceeds to the next step.

(5) Image recognition When a network is created as described above, image recognition can be performed based on this network. In this embodiment, the direction (angle) of each arc in the network is obtained (step S220). This is effective, for example, when a knitting machine or the like needs to know the longitudinal direction.

First, for the network formed as described above, the vector sum of arcs for each of the labels A, B, and C is calculated by the following equations (1), (2), and (3).

[0035]

[Equation 1]

[0036]

[Equation 2]

[0037]

[Equation 3]

From each vector sum, the direction of each vector can be obtained by the following equations (4), (5) and (6).

[0039]

[Equation 4]

[0040]

[Equation 5]

[0041]

[Equation 6]

When the direction of the longitudinal grain is known in advance, the angle closest to it can be determined as the longitudinal direction. In the case where the screen is repeatedly imaged, if the vertical direction is already known in the previous screen, the vertical direction can be calculated by comparing with the previous screen. Next, the obtained arc is displayed as an image in association with the stitch (step S230). FIG. 16 shows an example of the stitch arrangement structure obtained.

[0043]

According to the present invention, a network is created from a structure, and judgment is made from the structure of this network.
It is freed from parameters such as distance and angle between nodes. Therefore, the recognition rate can be improved even when it is deformed to some extent like a stitch, and thereby the directionality and the angle can be detected with high accuracy.

Further, according to the present invention, the image preprocessing (outline extraction, etc.) cannot be accurately performed, and the portion which should be originally surrounded by the outline is divided into two parts. Even if it happens, the original structure can be recognized by the relative positional relationship.

[Brief description of drawings]

FIG. 1 is a block diagram showing a basic configuration of the present invention.

FIG. 2 is an example of a hardware configuration diagram for implementing the present invention.

FIG. 3 is a flowchart showing an operation procedure of the present invention.

FIG. 4 is a diagram showing a node centroid table.

FIG. 5 is a diagram showing an adjacent point table.

FIG. 6 is a diagram showing an arc table.

FIG. 7 is a diagram showing a relationship between arcs and nodes according to the present embodiment.

FIG. 8 is a diagram illustrating a condition 1-1.

FIG. 9 is a diagram illustrating a condition 1-2.

FIG. 10 is a diagram illustrating a condition 1-3.

FIG. 11 is a diagram illustrating a condition 2-1.

FIG. 12 is a diagram illustrating a condition 2-2.

FIG. 13 is a diagram illustrating a condition 3-1.

FIG. 14 is a diagram illustrating a condition 3-2.

FIG. 15 is a diagram illustrating a condition 4;

FIG. 16 is a diagram showing an image display of an array structure of arcs and nodes in the stitch of this embodiment.

FIG. 17 is a diagram showing a method of photographing a knitted fabric.

FIG. 18 is a diagram showing stitches.

[Explanation of symbols]

 100 image input means 200 contour extraction means 300 barycentric position detection means 400 network setting start point determination means 500 network setting means 600 recognition means 700 output means 1 central processing unit (CPU) 2 image processor 3 image processing memory 4 lead Only memory (ROM) 5 Random access memory (RAM) 6 Input / output interface (I / O) 7 Bus 8 Imaging device 9 Keyboard 10 Display device

Claims (2)

[Claims] 1. A contour of each of the patterns is extracted from an image signal having a plurality of patterns arranged according to a certain rule, and barycentric positions within a range surrounded by the contours are detected, and each barycentric position is used as a node. , A line segment connecting an arbitrary center-of-gravity position and an adjacent center-of-gravity position as an arc, the network is sequentially expanded from a network setting start point based on a predetermined relative positional relationship condition based on the above-mentioned certain rule, and this network is expanded. An arrangement recognition method characterized by recognizing an array of images based on the. 2. An image inputting means, a contour line extracting means for extracting a contour of a pattern from an image signal inputted from the image inputting means, and a center of gravity for detecting and storing a barycentric position of a range surrounded by the contour. Position setting means, a center of gravity position as a node, a line segment connecting an arbitrary center of gravity position and its adjacent center of gravity position as an arc, and a network setting start for determining a predetermined node for setting a network of nodes and arcs The point determining means, the network setting means for sequentially expanding the network from the network setting start point based on a predetermined relative positional relationship condition, and setting the network within the range of the image signal, and the network setting means. A recognition means for recognizing the arrangement of images based on the recognition means; and an output means for outputting the result recognized by the recognition means. An image recognition device characterized by the following.