JP3557814B2 - Image inspection apparatus and method - Google Patents

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an image inspection apparatus and method for inspecting, collating, and sorting products using a digital image pattern matching technique.
[0002]
[Prior art]
In a field to which the pattern matching technology is applied, there is an inspection system that recognizes an image of a target article, performs matching with a reference known image, and determines the quality of the image. In such an inspection system, a target article is transported by a conveyor or the like, and a pattern such as a character, a symbol, or a pattern attached to the surface of the target article is captured using an imaging device, and the captured image and a known image registered in advance. The quality of the target article is determined based on the result of the matching.
[0003]
[Problems to be solved by the invention]
However, in the case where the number of target articles is small and large in number, it is necessary to manually set the pass / fail judgment parameters by manual operation, which requires a great deal of labor.
[0004]
Therefore, there are a method of registering a plurality of judgment parameters in advance and switching as appropriate, and a method of automatically extracting and registering judgment parameters necessary for matching from a real work by a learning method.
[0005]
However, in the former method, since the switching is performed manually, the labor is required, and the switching is performed subjectively by the operator. In addition, in the latter method, since the determination parameter can be extracted only after inspecting several target articles, there is an inconvenience that the method is not suitable for a target article of a small quantity and many kinds. If the determination parameters are not optimized, there is a disadvantage that a good product is erroneously determined as a defective product or a defective product is erroneously determined as a non-defective product, and sufficient matching accuracy cannot be obtained.
[0006]
The present invention has been made to solve such a conventional problem, and an object of the present invention is to provide an image inspection apparatus and a method capable of automatically adjusting and obtaining an optimum determination parameter.
[0007]
[Means for Solving the Problems]
The invention according to claim 1 of the present invention recognizes an image of a target article, and determines whether or not the image of the target article is within a predetermined allowable value with respect to a predetermined image area of a reference master image. determining an image inspection apparatus, the allowable value, the set for each of a plurality of small regions constituting the image area, said image area for the feature amount a calculated in each small region of the master image When the characteristic amount B obtained for each of the small areas after moving is within a predetermined value P1 with respect to the characteristic amount A, the characteristic amount B is set to P1. If not within P1 is that you and P2 corresponding to the difference between the feature amount B and the feature amount a and feature.
[0009]
In the image inspection method according to the second aspect of the present invention, an image area to be inspected is set in a master image, the image area is divided into a plurality of small areas, and a feature amount A is set for each small area. Then, the image area is moved, and a characteristic amount B is extracted for each small area after the movement. When the characteristic amount B is within a predetermined value P1 with respect to the characteristic amount A, an allowable value of the small area is changed. and P1, the when not within the predetermined value P1 is set to P2 corresponding the permissible value of the small region on a difference between the feature amount a and the feature amount B the feature amount B is relative to the feature amount a, the A master registration process of setting an allowable value for each small area, an image of a target article is captured as an inspection image, this inspection image is made to match the master image, and the inspection image is included in the inspection image in the same manner as the master image. Image area and a plurality of small areas Extracting D, and characterized by comprising a inspection process each feature amount D is examined whether or within the allowable value P1 or P2 of the corresponding small regions.
[0010]
According to the present invention, a simulation that takes into account variable elements predicted when matching a master image and an inspection image is automatically performed in advance, that is, an image area to be inspected is moved to make an erroneous determination The erroneous determination rate at the time of matching is reduced by automatically changing the allowable value for a small area having a high value.
[0011]
BEST MODE FOR CARRYING OUT THE INVENTION
FIG. 1 is a block diagram showing an embodiment of an image inspection apparatus according to the present invention. In FIG. 1, an image captured by a video camera 1 is converted into digital image data by an image input unit 2 and stored in an image memory 3. The image data stored in the image memory 3 is read by the image output unit 4, converted into an analog image signal, and displayed on a display device such as a video monitor 5.
[0012]
A CPU (Central Processing Unit) 6 functions as a control unit for controlling the entire device, and a program memory (ROM) 8 in which a program executed by the CPU 6 is stored via a system bus 7 including an address bus and a data bus. , A working memory (RAM) 9 for storing various data input from the outside, processing results in the CPU 6, and the like, a judgment parameter memory 10 for storing allowable values (judgment parameters) described later, and an image to be inspected. It is connected to an input unit 11 for setting an image area, dividing the image area into a plurality of small areas (blocks), and setting various parameters.
[0013]
The CPU 6 is connected to the image input unit 2, the image memory 3, and the image output unit 4 via the system bus 7. Further, a feature amount extraction unit 12 is connected between the image memory 3 and the system bus 7. The feature amount extraction unit 12 is for extracting a feature amount for each divided block, and uses, for example, a standard deviation value σ. The standard deviation value σ is defined as follows: In a density histogram in a certain image area, when the number of pixels for a density level i (i = 0 to 255) is Pi and the number of all images is N, the density average value μ is
[0014]
It becomes. From this, the density dispersion value V is
[0015]
It becomes. The standard deviation value σ from the density variance value V is
[0016]
It becomes.
[0017]
(Operation explanation)
Next, the operation of the present invention will be described. The present invention captures an image of an article serving as a reference for inspection as a master image, determines a feature amount of the master image, sets a determination parameter for each small area, and performs image registration of an object to be inspected. As an inspection image, comparing the feature amount of the master image with the feature amount of the inspection image, and determining the acceptability of the target article from the determination parameter for each small area.
[0018]
In the present embodiment, the image of the first target article of the production lot is captured as a master image, the images of the second and subsequent target articles are captured as inspection images, and the feature amounts of the master image and the inspection image are compared. Are compared to determine the acceptability of the second and subsequent target articles of the production lot.
[0019]
(Master registration process)
First, referring to the flowchart shown in FIG. 2 and the image pattern diagram shown in FIG. 3, an image of the first target article of a production lot is taken in as a master image serving as a reference for inspection, and a master for obtaining a feature amount of this master image is obtained. The procedure of the registration process will be described.
[0020]
First, initial settings are made (step S1, pattern PG1). In this initial setting, an image area (hereinafter, referred to as a window) WD to be inspected is set by the input unit 11, and the inside of the window WD is further divided into a plurality of blocks (a plurality of m × n pixels) BL.
[0021]
Next, the first target article of the production lot serving as a reference for inspection is imaged by the video camera 1 and taken into the image memory 3 as a master image (step S2, pattern PG2). At this time, the video camera 1 continuously captures the target article twice in the same place. This takes into account quantization errors and the like, and is stored in the two memory units 3a and 3b of the image memory 3 as a first master image Ma and a second master image Mb, respectively.
[0022]
Next, each window WD of the master images Ma and Mb is divided into four parts, and a misregistration correction mark is extracted from each divided area. The misregistration correction mark serves as a reference when performing alignment between the master image Ma and an inspection image TG described later.
[0023]
In the method of extracting the misregistration correction mark, the one having the largest feature amount in each of the four divided areas is set as a representative value, and two representative values are selected from the four representative values according to a predetermined priority order. The two are determined as position shift correction marks Ua and Ub, and the position data is stored in the RAM 9 (step S3, pattern PG3).
[0024]
For example, as shown in a pattern PG3, the inside of the window WD is divided into four regions WD1 to WD4, and a mark at a position having the largest feature amount has edge information, and two marks which are as far apart from each other as possible are selected. The priorities are, for example, WD1-WD3, WD2-WD4, WD1-WD2, WD2-WD3, WD3-WD4, WD4-WD1. In this embodiment, a black star in the upper left area WD1 of the image and a white star in the lower right area WD3 are selected as the displacement correction marks Ua and Ub.
[0025]
Next, a standard deviation value (hereinafter, referred to as a deviation value) σa as the feature amount A is extracted for every block BL of the window WD set in the master image Ma (step S4). Next, a determination parameter P1 according to the extracted deviation value σa is set and stored in the determination parameter memory 10 (step S5, pattern PG4). Each determination parameter P1 is a predetermined value set in advance according to the value of the deviation value σa regardless of the production lot. As a result, the judgment parameter memory 10 stores a judgment parameter P1 having a value corresponding to the deviation value σa of each block.
[0026]
Next, the window WD is moved by several pixels to the upper left on the image (step S6, pattern PG5), and the feature amount B of each block of the moved window is extracted as the deviation value σb (step S7), and the corresponding value before the movement is obtained. A comparison is made with the block deviation value σa (step S8).
[0027]
That is, whether the deviation value σb is within the determination parameter P1 with respect to the deviation value σa is compared for each block by the following equation,
σa−P1 <σb <σa + P1 (4)
The value of the determination parameter P1 of the block that satisfies the condition of Expression (4) is left unchanged (Step S9), and the value of the determination parameter P1 of the block that does not satisfy the condition of Expression (4) is changed (Step S10). The value of the new determination parameter P2 at this time is the difference between the feature value A (deviation value σa) and the feature value B (deviation value σb),
P2 = | σb−σa | + β (5)
It becomes. β is a parameter for adjusting the severity of the determination. This process is performed for all blocks (step S11).
[0028]
Next, the window WD is moved by several pixels in the other direction, and the processing of steps S7 to S11 is finally performed in four directions (up, down, left, and right) (step S12), and the determination parameter P1 or P2 for each block is determined.
[0029]
FIG. 4 shows four virtual inspection images Ia to Id obtained by moving the window WD up, down, left, and right. The image area indicated by the solid line is the window WD before the movement, and the image area indicated by the broken line is the virtual inspection images Ia to Id.
[0030]
In this example, assuming that the start position coordinates of the window WD are (x, y), FIG. 12A shows the virtual inspection image Ia of the start position coordinates (x−α, y−α), and FIG. The virtual inspection image Ib at (x + α, y-α), FIG. (C) is the virtual inspection image Ic at the start position coordinates (x−α, y + α), and the diagram (d) is the virtual inspection image at the start position coordinates (x + α, y + α). The image Id is shown.
[0031]
As a result, if the condition of Expression (4) is not satisfied in many blocks, it is determined that the initial setting needs to be changed, that is, the look-up table needs to be changed (step S13), and the value of the preset determination parameter P1 is determined . The lookup table indicating the relationship is changed by the input unit 11 (step S14), and the processes after step S6 are executed again.
[0032]
(Inspection processing)
Next, the procedure of the inspection process for determining the acceptability of the second and subsequent target articles in the production lot will be described with reference to the flowchart shown in FIG. 5 and the image pattern diagram shown in FIG. First, when the target article to be inspected is transported on the conveyor and reaches the inspection area, it is imaged by the video camera 1 and taken into the memory unit 3b of the image memory 3 as the inspection image TG (step T1, pattern QG1).
[0033]
Next, misregistration correction marks Va and Vb are extracted from the inspection image TG (step T2, pattern QG2). These displacement correction marks V and Vb correspond to the displacement correction marks Ua and Ub of the master image Ma, and are extracted by the same method as in step S3 described above.
[0034]
Then, the inspection image TG stored in the memory unit 3b of the image memory 3 is stored in the memory unit 3a so that the position error correction marks Va and Vb match the position error correction marks Ua and Ub. Position adjustment for moving the inspection image TG so as to match the image Ma is performed (step T3). This is for aligning the target articles which are transported out of alignment on the conveyor on the image.
[0035]
Next, a deviation value σd as a feature value D is extracted for each block BL of the inspection image TG (step T4). Then, it is checked whether the deviation value σd of each block of the inspection image TG is within the determination parameter P1 or P2 with respect to the deviation value σa of each corresponding block of the master image Ma (step T5, pattern Q3). Since the determination parameter P1 is set for each block and stored in the determination parameter memory 10, the deviation value σd of each block must satisfy the following equation.
[0036]
σa−P1 <σd <σa + P1 (6)
Or
σa−P2 <σd <σa + P2 (7)
This inspection is performed on all blocks of the inspection image TG (step T6), and when the inspection is completed, the inspection result is output to the video monitor 5 (step T7). If it is determined that the block is non-defective, the video monitor 5 displays the fact. If it is determined that the block is defective, the number of blocks that did not satisfy the condition of Expression (6) , its image, its position, and the like are displayed. This series of inspection processing is performed for all target articles (step T8), and when inspection of all articles is completed, the processing is ended.
[0037]
In the above-described embodiment, an image in four directions is assumed as the tentative inspection image. However, the present invention is not limited to this, and a simulation is performed by assuming an image in more directions (for example, eight directions). Is also good. Further, the feature amount is not limited to the standard deviation value, but includes an average density (brightness) in the block, a color difference, a normalized correlation, and the like.
[0038]
【The invention's effect】
According to the present invention, a simulation using a temporary inspection image is automatically executed in advance in consideration of a variable element predicted at the time of matching between a master image and an inspection image, and a block having a high erroneous determination element ( Since the determination parameter (allowable value) is automatically changed for the (small area), a manual operation becomes unnecessary, and the effect of reducing the erroneous determination rate of matching is obtained.
[Brief description of the drawings]
FIG. 1 is a block diagram of an image inspection apparatus according to an embodiment of the present invention.
FIG. 2 is a flowchart showing an operation procedure (master registration processing) of the present invention.
FIG. 3 is a diagram illustrating an image pattern of a master registration process.
4A to 4D are diagrams showing four virtual inspection images obtained by moving a window in a master image up, down, left, and right by several pixels.
FIG. 5 is a flowchart showing an operation procedure (inspection processing) of the present invention.
FIG. 6 is a diagram showing an image pattern of an inspection process.
[Explanation of symbols]
Reference Signs List 1 video camera 2 image input unit 3 image memory 4 image output unit 5 video monitor 6 CPU
7 System bus 8 Program memory (ROM)
9 Working memory (RAM)
10 judgment parameter memory 11 input unit 12 feature amount extraction unit WD window (image area)
BL block (small area)
Ma, Mb Master images Ua, Ub Misalignment correction marks Va, Vb Misalignment correction marks Ia to Id Temporary inspection image TG Inspection image A Feature amount of master image B Feature amount of virtual inspection image D Feature amount of inspection image σa Master image Standard deviation value σb of the virtual inspection image Standard deviation value σd of the inspection image Standard deviation values P1 and P2 Judgment parameters (allowable value)

Claims (2)

An image inspection apparatus that recognizes an image of a target article and determines whether the image of the target article is within a predetermined allowable value for a predetermined image area of a master image that is a reference,
The allowable value, the set for each of a plurality of small regions constituting the image area of the master image, the respective small regions wherein after moving the image area for the feature amount A calculated in each small region When the characteristic amount B obtained for each of the characteristic amounts A is within the predetermined value P1 with respect to the characteristic amount A, the characteristic amount is set to P1. When the characteristic amount B is not within the predetermined value P1 with respect to the characteristic amount A, the characteristic amount is set as P1. image inspection device according to feature that you and P2 corresponding to the difference between the B and the feature amount a. An image area to be inspected is set in the master image, the image area is divided into a plurality of small areas, a feature amount A is extracted for each small area, and the image area is moved to move each small area. A feature value B is extracted for each region. When the feature value B is within a predetermined value P1 with respect to the feature value A, the allowable value of the small area is set to P1. When the value is not within the predetermined value P1, a master registration process of setting an allowable value for each small region as P2 corresponding to the difference between the feature amount B and the feature amount A as an allowable value of the small region;
An image of a target article is captured as an inspection image, this inspection image is matched with the master image, and an image area to be inspected and a plurality of small areas are set in the inspection image in the same manner as the master image. An inspection process of extracting a characteristic amount D for each region and inspecting whether each characteristic amount D is within the allowable value P1 or P2 of the corresponding small region;
Tona images inspecting method comprising Rukoto.