JPH07120418B2 - Pattern recognition method - Google Patents

Description

The present invention relates to a pattern recognition method suitable for use in a semiconductor manufacturing apparatus or the like.

[Prior Art] When wire bonding a semiconductor chip bonding portion and a frame bonding portion with a wire bonding device, in order to maintain the chip and the frame in an optimum positional relationship,
It is necessary to detect the displacement of the chip. This displacement is detected by the pattern recognition device.

In the conventional pattern recognition method, a captured image obtained by shooting a chip-shaped detection pattern with a television camera is binarized by a reference binarization level by an image processing circuit, and the binarized image is scanned. Then, each part of the binarized image is compared with the reference pattern registered in the pattern setting circuit in advance, and the degree of matching between each part of the binarized image and the reference pattern is calculated by the correlation circuit. The position that best matches the reference pattern is found in the binarized image, and the displacement from the reference position is calculated.

[Problems to be Solved by the Invention] However, in the conventional pattern recognition method, when the reference pattern is registered in the pattern setting circuit in advance, the reference binarization level and the low-level image part and the high level in the reference pattern are high. The operator decides the area ratio setting of the level image part. In such a case, the reference binarization level or the area ratio of the low-level image portion and the high-level image portion in the reference pattern deviates from the optimum state as described below, and the pattern recognition accuracy, that is, the object The position detection accuracy includes variations based on individual differences among operators.

The present invention, when defining a reference pattern, automatically optimizes the reference binarization level and the area ratio of the low-level image portion and the high-level image portion in the reference pattern to improve the position detection accuracy of the subject. To aim.

[Means for Solving Problems] According to the present invention, an imaged image of a subject is binarized at a reference binarization level, and the binarized image is compared with a predetermined reference pattern. In the pattern recognition method for detecting the position of the subject based on the degree of matching between the binarized image and the reference pattern by this comparison, (A) the reference image to be captured as the reference pattern in determining the reference binarization level. Is binarized to change the binarization level used as the reference binarized image, and the binarization level that minimizes the change in the reference binarized image obtained in response to the change in the binarization level is set as the reference. In addition to the binarization level, (B) the size of the outer frame range of the reference binarized image binarized at the reference binarization level is changed, and the image surrounded by the outer frame for each change Low level image area and high level in area The ratio occupied by each image part is obtained, and the reference binarized image within the outer frame range where the ratio is approximately 50% is used as the optimum reference pattern.

[Operation] The reference binarization level used when the reference pattern is predetermined must be capable of stably binarizing the captured image when the captured image of the subject is subsequently binarized.
FIG. 4 (A) is a schematic diagram showing the subject 1, and FIG. 4 (B) is a schematic diagram showing a captured image (video signal) 3 obtained along the scanning line 2 of the subject 1. The captured image 3 changes from a state indicated by a solid line to a state indicated by a one-dot chain line or a state indicated by a two-dot chain line due to, for example, changes in the amount of illumination light, variations in reflectance of the subject 1, and the like. Therefore, the reference binarization level has a small change in the binarized image obtained by binarizing the captured image 3 even if the captured image 3 changes as described above. It is necessary to set to the optimum level shown in.

Thus, as in the present invention, when the reference image to be captured as the reference pattern is binarized, the binarization level is changed, and the change (low level of the reference binarized image obtained at that time is changed. If the binarization level that minimizes the change in the area ratio between the image portion and the high-level image portion is used as the reference binarization level, the optimization of the reference binarization level can be realized.

Further, in the position detection operation of the subject based on the degree of agreement between the binary image of the subject and the reference pattern, the obtained degree of agreement is compared with the reference degree of matching to determine whether or not the detection is successful. This standard matching degree does not take a very large value because there are changes in the detection environment and the state of the subject, and empirically, for example,
70-80%. Therefore, under the condition that the area ratio of the low-level image part and the high-level image part in the reference pattern is set to 20% to 80%, all the original binary images of the subject to be compared with the reference pattern are high-level. If it is an image part, there is a 80% matching degree at that time, which may cause erroneous detection.

Then, as in the present invention, the size of the outer frame range of the reference binarized image is changed, and the ratio of the low level image part to the high level image part is approximately 50% within the outer frame range. Criterion 2
When the binarized image is used as the reference pattern, the area ratio in the reference pattern can be optimized.

Therefore, according to the present invention, when the reference pattern is determined, the reference 2 that can obtain a stable binarized image with little change in response to a change in the shooting condition such as a change in the amount of illumination light.
Quantization level The size of the outer frame that optimizes the area ratio between the low-level image part and the high-level image part is automatically set, preventing false detection and improving the position detection accuracy of the subject. it can.

[Embodiment] FIG. 1 is a block diagram showing a pattern recognition apparatus used for implementing the present invention, FIG. 2 is a schematic diagram showing a method of determining a reference binarization level, and FIG. 3 is an outer frame range of a reference pattern. It is a schematic diagram which shows the state which changes.

In FIG. 1, 11 is a television camera, 12 is an image processing circuit, 13 is a computer, 14 is a D / A conversion circuit, 15 is an image memory, 16 is a correlation circuit, 17 is a pattern setting circuit, and 18 is a mask setting circuit. is there.

First, a method of detecting the position of the subject by this pattern recognition device will be described.

The television camera 11 captures an image of a detection pattern on a subject such as a semiconductor chip and sends the captured image (video signal) to the image processing circuit 12. The image processing circuit 12 is a computer
The reference binarization level registered in 13 is used for the D / A conversion circuit 1
The image pickup image is given via 4 and is binarized by the reference binarization level. The binarized image is represented by 0 (high level image part: white image part) or 1 (low level image part: black image part) according to the contrast of the picked-up image. For example, the X direction is 128 bits and the Y direction is It is specified as 128 bits and stored in the image memory 15.

The binarized image is scanned as follows.

That is, the correlation circuit 16 reads the data A at the coordinates (0,0) of the binary image stored in the image memory 15.
The data A of the coordinates (0,0) is, for example, binarized data at each point within 32 bits × 32 bits. The correlation circuit 16 registers each point of the data A input as coordinates (0,0) and the pattern setting circuit 17 with, for example, 32 registered in advance.
Bits x 32 bits are compared with corresponding points of the reference pattern B, and whether or not they match, that is, the degree of matching is calculated,
To memory. At this time, if the obtained degree of matching is lower than a predetermined reference degree of matching, it is determined as a detection failure and the calculation result of the degree of matching is ignored and not stored.

In the calculation of the matching degree, the level value (0 or 1) of each point of the data A and the level value (0 of each point of the reference pattern B are set for optimization of the reference pattern described in detail later.
Alternatively, the exclusive OR of 1) is obtained, and the negative form of this exclusive OR is ANDed with the level value (0 or 1) of each point of the reference mask degree C2 set in the mask setting circuit 18. By doing so, the degree of agreement between the data A and the optimum reference pattern is calculated.

FIG. 5A to FIG. 5E specifically show the procedure for calculating the degree of coincidence. The figure shows 4 bits x
This is an example of 4 bits. 5A shows the data A, FIG. 5B shows the reference pattern B, FIG. 5C shows the reference mask degree C2, and FIG. 5D shows the data A and the reference. A negative form of the exclusive OR of the pattern B is shown, and FIG. 5 (E) shows a logical product of the negative form of the exclusive OR and the reference mask degree C2. The shaded areas in FIGS. 5A, 5B, 5D, and 5E represent 1 as the low-level image area, and the non-hatched areas represent 0 as the high-level image area. The shaded area in FIG. 5 (C) represents 1 as an effective area described later, and the non-shaded area represents 0 as an invalid area. Here, the reference pattern B
As will be described later in detail, the area ratio of the low level image portion and the high level image portion is set to 50% to 50% by the reference mask degree C2.

When the comparison of the data A at the coordinates (0,0) is completed, the correlation circuit 16 reads the data A at the coordinates (1,0) consisting of 32 bits × 32 bits shifted by 1 bit in the X direction, and the coordinates ( The data A of 1,0) is compared with the reference pattern B. In this way, the correlation circuit 16 sequentially scans each data from the coordinates (0,0) to the coordinates (95,95) and calculates the degree of coincidence. When the correlation circuit 16 finishes the calculation of the matching degree over the entire range of the binarized image stored in the image memory 15, it selects the position with the largest matching degree stored in itself and selects the position from the reference position. The position shift (x, y) is obtained, and the position of the subject is detected by this.

By the way, the reference binarization level registered in the computer 13 is set when the reference pattern is determined in advance. This reference binarization level is used for binarizing the captured image of the subject in the image processing circuit 12. The captured image, which may change depending on the shooting conditions, etc.
It must be quantifiable.

Therefore, in this embodiment, when the reference image 20 as shown in FIG. 2 to be captured as the reference pattern is binarized in the image processing circuit 12, the binary level given by the computer 13 is Δl. With a level difference of 2
The binarization level a and the binarization level b are used, and the levels a and b are gradually moved and changed in the range of l1 to l2.
Thereby, the reference binarized image obtained for the level a (for example, the area ratio of the high level image portion is 40%, the area ratio of the low level image portion is 60%) and the reference binarized image obtained for the level b (for example, the high level image portion). A and b when the change in the area ratio of the image part is 30% and the area ratio of the low level image part is 70%) (the change in the area ratio of the high level image part and the low level image part: eg 10%) is minimized. The value in the middle of is the optimum reference binarization level.

That is, according to this embodiment, since the reference binarization level is automatically optimized when the reference pattern is determined, the captured image of the subject is always stably binarized by the reference binarization level, The position detection accuracy of the subject can be improved.

Further, in the correlation circuit 16, when calculating the matching degree between the data A read from the binarized image of the subject and the reference pattern B, as described above, the obtained matching degree is used as the reference matching. The pass / fail judgment of the detection is made in comparison with the degree. This standard matching degree cannot be a very large value because there is a change in the detection environment or the state of the subject, and empirically, for example, 70-80
%. Therefore, under the condition that the area ratio of the high-level image part and the low-level image part in the reference pattern is set to 80% to 20%, the original 2 of the subject to be compared with the reference pattern 2
If all of the binarized images are high-level image parts, there is a 80% matching degree at that point, which may result in erroneous detection.

Therefore, in this embodiment, the reference image to be captured as the reference pattern is binarized by the image processing circuit 12 at the reference binarization level to obtain the reference binarized image, and then the reference binarized image is obtained. The outer frame range is gradually changed from the initial outer frame indicated by Z1 in FIG. 3, and the low level image portion (L) and the high level image portion (H) within the image area surrounded by the outer frame are changed for each change. The occupying ratio is calculated, and the reference binarized image surrounded by the outer frame range Z2 having the ratio of about 50% is used as the optimum reference pattern for calculation of the degree of matching by the correlation circuit 16.

A specific method of changing the outer frame range of the reference binarized image is to set the outer peripheral part of the initial outer frame Z1 of, for example, 32 bits × 32 bits to 0 (invalid area) and the central part to 1 by the mask setting circuit 18. The masking degree C of the outer frame Z that has been changed to the (effective area) is used. That is, the correlation circuit 16 sets the level value (0 or 1) of each point of the data A read from the reference binarized image obtained by the image processing circuit 12 and, for example, 32 bits × preliminarily set in the pattern setting circuit 17. After taking the exclusive OR of the level value (1) of each point of the whole surface leveling pattern B0 in which all the 32 bit points are the high level image part, the negative form of this exclusive OR and the mask degree The logical product of the level values of each point of C (0 on the outer side of the outer frame Z and 1 on the inner side) is calculated. As a result, the correlation circuit 16 can calculate the ratio of each of the high-level image portion and the low-level image portion in the image area surrounded by Z (zero) (ineffective area) in the fixed area of the outer portion. . Therefore, if the range of the outer part of the outer frame Z that is made 0 (ineffective region) is gradually expanded or contracted to change,
The mask degree of the outer frame Z2 that is approximately 50% of the above area ratio, that is, the reference mask degree C2 can be determined. That is, in this embodiment, the reference pattern set in the pattern setting circuit 17 is optimized by giving the reference mask degree C2 set in the mask setting circuit 18.

However, in the practice of the present invention, the reference mask degree C2
It is also possible to fix the reference pattern given with as the optimum reference pattern and set the optimum reference pattern in the pattern setting circuit 17.

That is, according to this embodiment, when the reference pattern is predetermined, the area ratio of the low-level image portion and the high-level image portion in the reference pattern can be automatically set to 50% to 50%, and the reference pattern is Since the optimization is performed, the pass / fail determination of the detection is ensured, and the position detection accuracy of the subject can be improved.

EFFECTS OF THE INVENTION As described above, according to the present invention, when the reference pattern is determined, the reference binarization level and the area ratio of the low-level image portion and the high-level image portion in the reference pattern are automatically optimized. As a result, erroneous inspection can be prevented and the position detection system of the subject can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a pattern recognition apparatus used for implementing the present invention, FIG. 2 is a schematic diagram showing a method of determining a reference binarization level, and FIG. 3 is a state in which an outer frame range of a reference pattern is changed. 4A and 4B are schematic diagrams showing the subject and its captured image, and FIGS. 5A to 5E.
[Fig. 3] is a schematic diagram showing a procedure for calculating the degree of agreement. 12 ... Image processing circuit, 15 ... Image memory, 16 ... Correlation circuit, 17 ... Pattern setting circuit, 18 ... Mask setting circuit,
20: standard image, A: binary image data, B: standard pattern, C2: standard mask degree, L: low level image part, H: high level image part, Z1, Z2. Outer frame, a, b ...
… Binarization level.

Claims (1)

[Claims] 1. An imaged image of a subject is set to 2 at a reference binarization level.
Pattern recognition in which a binarized image is binarized, the binarized image is compared with a predetermined reference pattern, and the position of the subject is detected based on the degree of matching between the binarized image and the reference pattern by this comparison. In the method, (A) in determining the reference binarization level, the reference image to be captured as a reference pattern is binarized to change the binarization level to be the reference binarization image, and the change of the binarization level. The binarization level that minimizes the change in the reference binarized image obtained corresponding to is set as the reference binarization level, and (B) the reference binarization binarized at the reference binarization level. The size of the outer frame range of the image is changed, and the ratio of each of the low level image part and the high level image part in the image area surrounded by the outer frame is calculated for each change, and the ratio becomes about 50%. Standard 2 values within the outer frame range Pattern recognition method which is characterized in that an image with optimum reference pattern