JP4620435B2 - Pattern matching method - Google Patents

Description

  The present invention relates to a pattern matching method that is applied to an inspection apparatus that automatically determines a quality of, for example, a medal or the like, which is automatically produced in large quantities such as a metal stamped product or a resin molded product and has unevenness and shading.

Conventionally, as this type of inspection, a pattern is discriminated based on the degree of matching of an unknown pattern with a standard pattern (see Patent Document 1), or a shape contour is tracked and extracted by connectivity analysis (see Patent Document 2). Etc.).
Japanese Patent Laid-Open No. 04-024749 (page 1-2, FIG. 4) Japanese Patent Laid-Open No. 04-049150 (page 4-6, FIG. 1)

However, each of the above methods inspects the characteristics of characters, figures, etc. recorded on a flat surface, and the recognition algorithm itself is used to inspect objects with both unevenness and shading, such as medals. Even though it can be partially used, there is a problem that an efficient and highly accurate inspection cannot be performed.
Therefore, the subject of this invention is enabling it to test | inspect the quality of the target object which has both unevenness | corrugation and lightness efficiently and with high precision.

In order to solve such a problem, in the invention of claim 1, an object having unevenness and shading is imaged, the captured image is divided into pixel units, a density value is obtained for each pixel, and an object is obtained by pattern matching. In pattern matching method to inspect things,
The area including the object is divided into a plurality of small areas including an outer peripheral area, a pattern area, a plain area, and a character area, and a similarity and a density ratio with a master pixel serving as a reference are calculated for each small area, The density dispersion value is calculated, and whether the similarity is equal to or greater than a value that takes a margin into consideration for a predetermined set value, and each of the density ratio and density variance is within a range that takes a margin into consideration for a predetermined upper and lower limit value, respectively. whether by have rows that quality determination, the similarity of the set value, together with the concentration ratio and the upper limit value and the margin of concentration dispersion is gradually updated by additional learning, the solid area slightly containing a pattern Is characterized in that this region is moved to a position where the pattern becomes a certain value or less, and the various operations are performed with the moved small region as a new small region .

According to this invention,
1) Compared to binarization, it is less susceptible to variations in shade on the surface of the inspection object.
2) The influence of fluctuation due to the brightness of the background is small.
3) Easy to follow position fluctuation and rotation fluctuation.
Therefore, there is an advantage that high-precision inspection is possible.

  FIG. 1 is a block diagram showing an apparatus to which the present invention is applied. This is an example in which the image processing apparatuses 1A and 1B are provided for the front side and the rear side of a medal or the like. Correspondingly, cameras 2A and 2B, illuminators 3A and 3B, illumination power sources E1 and E2, etc. are provided. It has been. Furthermore, an LCD monitor 7B is provided for rear display, a CRT monitor 7A is provided for front and rear display, and display switching between the front and rear is performed by the video switch 6. Distribution of the image data of the front surface and the rear surface is performed by the distributor 5. Reference numeral 8 denotes a video printer for printing video. Reference numeral 4 denotes a sequencer that controls the operation of the mechanical portion, and a data processing apparatus connected to the image processing apparatuses 1A and 1B is configured by a PC (personal computer: personal computer) 9, a monitor 10, a printer 11, and the like. Yes.

FIG. 2 is a flowchart for explaining the embodiment of the present invention.
First, an image is captured through the camera (step S1). The image processing apparatus divides the captured image into, for example, 480 × 512 pixels and binarizes each pixel. In step S2, it is determined whether the medal or the like is front or back. This is because the reference master differs depending on whether it is front or back. When the front or back is determined, a pixel for positioning the image is searched, and the position and angle are corrected so as to coincide with the master image (dictionary image) (steps S3 and S4).

  Next, in step S5, the determination target area is divided into small areas of about 20 × 20 pixels (including determination). Each of these small areas includes an outer peripheral part (peripheral area) such as a medal, a part with a pattern (pattern area), a part without a pattern or character (plain area), and a character part (designated area) for which a fine determination is to be made. It is classified into crab. Then, for each small region, various calculations in steps S8, 9, and 10 are performed. Before that, the outer diameter is measured (step S6), and for those with holes, the hole diameter is measured (step S7). It is.

The similarity calculation (concentration normalized correlation coefficient: similarity 0 to 100%) is performed by the following equation (1).
Similarity (concentration normalized correlation coefficient) =
Cross correlation coefficient between master image and inspection object image / autocorrelation coefficient between master image and inspection object image = {N · Σ (B · G) −ΣB · ΣG} / [{N · ΣB ² − (ΣB) ² } · {N · ΣG ² − (ΣG) ² } ^1/2 (1)
N: number of pixels in the small area, G: pixel density (0 to 255 gradations) in the small area of the master image,
B: Pixel density in a small area of the inspection target image (0 to 255 gradations)
It is.

The variance value in step S9 is obtained by the following equation (2), for example.
Dispersion value = [{ΣB ² − (ΣB) ² / N} / (N−1)] ^1/2 (2)
Note that N: the number of pixels in the small area, and B: the pixel density (0 to 255 gradations) in the small area of the image to be inspected are the same as in the case of equation (1).

In step S10, a density value (density ratio) indicating the degree of lightness and darkness compared with the master in the target small region is calculated.
Image density value = K / A / B
here,
A: dispersion value / density average value of corresponding small area of target image B: dispersion value / density average value of corresponding small area of master K: coefficient

Whether the product is good or defective is determined based on the similarity, dispersion, and concentration obtained as described above.
Step S11 is a similarity determination step. Here, for example, if it is equal to or greater than (master (setting) value−margin (margin)), a non-defective product is determined.
For the dispersion and density determination in steps S12 and S13, for example, a non-defective product is determined if it is within the range of (lower limit value−margin) to (upper limit value−margin), and a defective product is determined if exceeding the range.
Further, in step S14, the determination is made for all the small areas, and the non-defective product is determined only when all the small areas are OK.

  The master image captures and captures an appropriate number of normal products (basic master), but thereafter, the apparatus automatically calculates a determination allowable value and updates the master. Further, when the correct answer rate does not improve, it is desirable to perform further additional learning. In addition, when the master is updated, for an area that slightly includes the pattern area, the corresponding area is shifted in any of the front, rear, left, and right directions, and the position where the pattern becomes a certain value or less is set as a new small area. Assuming that, after that, various operations are performed in a small area at that position, thereby reducing the instability of the determination.

  In addition, in order to efficiently extract both unevenness and shading of medals etc., it is necessary to devise a certain way in the lighting method, but this point will be submitted separately. In the above description, the merit / defective inspection of medals has been mainly described. However, it goes without saying that the present invention can be similarly applied to coins.

Block diagram showing an apparatus to which the present invention is applied Flowchart showing an embodiment of the present invention

Explanation of symbols

DESCRIPTION OF SYMBOLS 1A, 1B ... Image processing apparatus, 2A, 2B ... Camera, 3A, 3B ... Illuminator, 4 ... Mechanical part sequencer, 5 ... Distributor, 6 ... Video switch, 7A, 7B, 10 ... Monitor, 8 ... Video printer, 9 ... PC (personal computer), 11 ... printer, E1, E2 ... illumination power supply.

Claims (1)

In a pattern matching method that images an object with unevenness and shading, divides the captured image into pixel units, obtains a density value for each pixel, and inspects the object by pattern matching,
The area including the object is divided into a plurality of small areas including an outer peripheral area, a pattern area, a plain area, and a character area, and a similarity and a density ratio with a master pixel serving as a reference are calculated for each small area, The density variance value is calculated, and whether the similarity is equal to or greater than a value that takes a margin into consideration for a predetermined set value, and each of the density ratio and density variance is within a range that takes a margin into consideration for the upper and lower limits. whether by have rows that quality determination, the similarity of the set value, together with the concentration ratio and the upper limit value and the margin of concentration dispersion is gradually updated by additional learning, the solid area slightly containing a pattern Is a pattern matching method characterized in that this region is moved to a position where the pattern becomes a certain value or less, and the various operations are performed using the moved small region as a new small region .

Images