JPS61213753A - System for detecting flaw on surface of object having regular reflection characteristic - Google Patents

Abstract

PURPOSE:To make it possible to detect the presence of the flaw on the surface of mirror surface matter of a flaw such as gentle bending, by making it possible to adapt an illuminance difference stereoscopic system to an object having regular reflection characteristics. CONSTITUTION:An image data input apparatus 10 is equipped with at least three light sources P1-P3, an irregular reflective surface 11 and a TV camera 12 and matter 13 is arranged in said apparatus 10. At first, known shape matter has regular reflection characteristics is arranged in the image data input apparatus 10 and a reference table expressing the relation between the set of an observed density value and the direction of the minute surface element of the surface of the aforementioned matter 13 is formed and held to a holding means. Next, with respect to matter to be inspected having regular reflection characteristics, the reference table holding means is referred by the set of the observed density value to extract the direction of the minute surface element of the surface of the object to be inspected. The flaw on the surface of the matter is detected on the basis of the extracted direction of the minute surface element.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a defect detection device for detecting defects on the surface of an object having specular reflection characteristics, such as a metal surface.

In particular, the present invention relates to a method for detecting defects on the surface of objects that can detect not only local defects such as scratches but also global defects such as gentle bends and twists.

For example, mechanical parts whose surfaces are plated with metal.

Conventionally, when detecting the presence or absence of scratches on the surface of an object that has specular characteristics, that is, specular reflection characteristics, or when the angle of an inclined surface is different from a standard one, these are usually checked visually.

By the way, in the case of such visual inspection, there are omissions in detection, so attempts have been made to automatically detect such defects using image processing technology in a computer.

[Conventional technology]

By the way, when detecting defects such as scratches using image processing technology in a computer, a grayscale image of the surface of the object is stored in a memory, and then binarized.
There is a method of detecting the presence or absence of scratches from the geometrical shape of the boundary of "1" or "0", and edge detection is performed by differentiating the density value.

There is a method of detecting scratches based on the edge strength.

However, all of these methods detect flaws mainly from lines such as contour lines, and it is not only difficult to set the darkness value, but also difficult to detect global defects such as gentle bends. There were problems such as not being able to.

By the way, for an object whose surface is diffusely reflective, the inclination of the object surface, that is, the direction (normal vector) of the microsurface elements on the object surface can be determined by a method called photometric stereo. This method uses rReflectance Map
Techniques for Analyzing
5surface Defects in MetaI
CastingsJ Robert J, Woodh
aa+ June 1978＾rtificial I
intelligence Laboratory M,
1. It has been published in T. The outline will be explained with reference to FIG.

As shown in FIG. 11(a), an object 1, such as a sphere, whose surface slope is known in advance is prepared. The surface of this object 1 has diffuse reflection characteristics.

Three light sources PI, P2. Place P3. And this light source is PI, P2. They light up individually in the order of P3. As a result, from TV camera 2, first
1 Image data as shown at dl in FIG. 1(b) is obtained, and then image data d2. ct3 is obtained. At this time, object 1
is a sphere, and the direction of each micro-area element is known, so that grayscale data for each micro-area element in one image data d1 to d3 can be obtained.

Therefore, based on the image data d1 to d3, by finding a combination of shading data for each microscopic surface element and using this as a reference table, an object whose inclination is unknown can be detected from the light sources P+, P2, .
By accessing the reference table using a combination of data for each micro-surface element obtained when irradiating the object individually in P3, the inclination of the unknown object for each micro-surface element can be obtained.

In other words, the direction (normal vector) of the microsurface element on the object surface
can be determined from multiple images observed from the same viewpoint under different lighting conditions. As mentioned above, this method is called photometric stereo, and the nine reflectance distribution maps (Re
reflectance Map),
The slope can be determined using the look-up table.

Generally, the apparent brightness on the surface of an object depends on four factors: the direction of the light source, the direction of the observer, the direction of the surface element, and the material of the surface. If the direction of the light source, the direction of the observer, and the material are known, the apparent brightness E is determined by the direction of the surface elements as the dependent variable.

It can be expressed as E=R(p, q). Here, (p, q) is the direction of the surface element expressed in a coordinate system based on the observer.

A plot of this E on the (p, q) plane is called a reflectance distribution map. It should be noted that the distribution is particularly dependent on the direction of the light source.

Suppose that the pixels of an image taken under a certain light source have an apparent brightness of E+. What is the orientation of the surface element on the subject that corresponds to this pixel?

E+=R+ (p, q)...-7
- Exists on a trajectory on the (p, q) plane that satisfies (1).

Suppose we get a brightness of E2 from the same pixel under the second light source, (p, q) is further.

E2 = R2(1), q)
It can be seen that -------42) is satisfied.

By the way, the positions of the subject and the television camera have not changed when capturing the first and second images.

The same pixel can be represented as the same position on the (p, q) plane. Therefore, the intersection of the above equations (1) and (2) becomes the direction of the surface element.

Furthermore, in reality, due to the nonlinearity of both equations, an image under a third light source is required.

In photometric stereo, there is no parallax between each image.

Matching can be done quickly. Lookup tables are often used to further increase speed. For this reason, the distances between the observer and the subject, as well as between the light source and the subject, are set to be sufficiently far compared to the size of the object. Under this assumption, the directions of the light source and the observer are considered constant in each plane element of the object. Therefore, the same reflectance distribution map can be used. In other words, the relationship between the three sets of brightness data and the orientation of the surface elements is unchanged depending on the position on the screen. Therefore, the relationship between the triplet (when using three light sources) of the brightness (shading value of the image data) and the orientation of the surface element is calculated in advance, and the relationship between the orientation of the surface element is stored as a reference table that allows you to refer to the orientation of the surface element from the triplet. Just leave it there.

[Problem that the invention seeks to solve]

By the way, in the conventional photometric stereo as mentioned above.

There is a problem in that this method cannot be used when dealing with objects that have mirror surfaces, which are particularly important in industrial applications, that is, objects that have significant specular reflection characteristics. In other words, because the object surface has specular reflection characteristics, 1. For example, when a point light source is used, the orientation of most surface elements on the object surface does not reflect light in the observation direction, and the observed image data is a collection of almost zero density levels. There is a problem that.

[Means for solving problems]

The present invention makes it possible to use the above-mentioned photometric stereo method even for objects having specular reflection characteristics, thereby extracting the direction of surface elements on the surface of the object. For this purpose, the present invention provides a method for detecting defects on the surface of an object having specular reflection characteristics.

at least three light sources, a diffused reflection surface that irradiates a target object with light by reflecting light from the light sources, an image input means for observing the target object, and each light source point of a known-shaped object having specular reflection characteristics. (3) a reference table creating means for creating a reference table expressing the relationship between a set of grayscale values observed under a light and the orientation of the microscopic plane elements on the surface of the object; and (3) a reference table holding means for holding the created reference table. , extraction means for extracting the orientation of the micro-area elements on the surface of the object to be inspected by searching the reference table holding means with a set of grayscale values observed under each light source point light of the object to be inspected having specular reflection characteristics; , detecting defects on the surface of the object to be inspected based on the output of the extraction means.

C effect) In the present invention, it is possible to obtain a set of gray values of the object to be inspected using the diffuse reflection surface, and based on this, discontinuities in the orientation of microscopic surface elements can be extracted and local defects such as scratches can be detected. can be detected. In addition, by using a model to extract the orientation of the microscopic surface elements in advance and comparing the orientation of the microscopic surface elements of the object to be inspected with that of this model, global defects such as gentle bends can be detected. I can do it.

〔Example〕

Before describing the present invention in detail based on one embodiment.

The outline will be explained based on FIGS. 1 to 3.

Figure 1 (al is a human-powered image data device for inputting image data in the present invention, and bl is a human-powered image data device for inputting image data in the present invention, and bl shows the center position and radius of an object such as a sphere for which each surface element is known when creating a lookup table. Flowchart for searching.

(C) and (d) are explanatory diagrams when extracting the orientation of surface elements.

(8) is a pro-nk diagram for creating a reference table and outputting the orientation of surface elements in the present invention, FIG. 2 is an explanatory diagram of the state of creating a reference table, and FIG. This is an example of the extraction result of the raw orientation.

In addition, in FIG. 3, the lower left diagram represents whether or not the direction of the surface element has been determined. The areas where the direction of the surface element was determined are shown in black, and the areas where the direction could not be determined are shown in white.

This is due to the display of the direction of the 9-sided element, so whether "・" is the direction of the actually calculated area element (the direction of the area element facing directly above), or
This is to distinguish whether it is a part that could not be found.

Find the direction of the surface element. It's not wanted.

This is because when creating a reference table, there are physically impossible combinations of three light sources, and the contents of the reference table are not necessarily all filled in. Therefore, in fact, as the contents of the reference table.

Keep information (1 bit) as to whether or not the contents of the reference table are actually meaningful combinations.

When looking up a reference table, the information determines whether or not the direction of a surface element has been found.

The lower left figure displays this information. Note that the meaning of this lower left diagram is the same in other diagrams (FIGS. 7, 8, etc.) to be described later.

In FIG. 1, lO is an image data input device, and a diffused reflection surface 11. TV left camera 2. Light sources P1 to P3 are installed 2 and an object 13 is placed therein. 14 is a center position/radius calculation circuit that calculates the center position and radius of an object, and 15 is a correction circuit that adjusts the light irradiated to the object 13 and the light input from the object 13 to the TV left camera 2. It is necessary that the rays be parallel, but in reality the object 13 and the TV
Since the distance between the left camera 2 and the diffused reflection surface 11 cannot be kept large, the brightness is corrected, and the correction amount for each image data is obtained from the correction coefficient table 16 and correction is performed using this. A normalization circuit that includes light sources PI and P2.
．． In order to remove the influence of P3, the respective maximum values DIlax+, Dmax2, D of memory el, e2.63 are set.
A device that detects tmax3 and calculates the quotient of data in each memory el to c3 by their respective maximum values, 18
is a maximum value search circuit, and the memories el, e2 . Each maximum value Omax% in e3, Dsax2, On
+ax3 is searched, 19 is a direction determination circuit, and in the case of "reference table creation mode" which creates a reference table using an object such as a sphere whose inclination of each pixel is known in advance, the center position radius The position of the surface element of the image data is calculated based on the data sent from the calculation circuit 14, and based on this position, the slope table 20 is accessed to obtain its normal vector L, and the memory M+, M2 . Enter the set of three density values normalized from M3 and this normal vector L in the reference table, and also detect the orientation of the surface elements of the object 23 to be inspected.
In the case of "Surface element orientation detection mode", the reference table 21 is accessed using the set of three density values for each surface element as a key, the corresponding normal vector, that is, the orientation of the surface element is determined and output. This is output to the output data section 22 as data.

Next, the creation of a reference table and the creation of a surface area of an object to be inspected will be explained.

+1) Create a reference table First, an object 13 that has the same reflection characteristics as the object to be inspected (if the object to be inspected has specular characteristics, it also has specular characteristics) and has a known shape (usually spherical) is used as the light source P+
, P2. Place it under P3. At this time, first, the three light sources P
Turn on I-P 3.

The object 13 is illuminated by the diffuse reflection surface 11, but its position is appropriately determined using an auxiliary light source as necessary so that no hidden lines are included in the image data of the object 13 in the TV left camera 2. In this way, TV left camera 2
The image data of the spherical object 13 is held in the memory mQ, and its center position and radius are calculated by the center position radius calculation circuit ■4
The contour of the spherical object 1'3 is thus obtained.

Next, turn on the light sources one by one. As a result, the image data when the light source P+ is turned on is held in the memory m1, the image data when the light source P2 is turned on is held in the memory m2, and the image data when the light source P3 is turned on is held in the memory m3.
will be held. The image data held in these memories m1 to m3 is corrected in the correction circuit 15 based on the correction amount obtained from the correction coefficient table 16, and then stored in the memory 61. e2. Retained in e3.

Then, the maximum value search circuit 18 searches the memories et, e2.
Maximum value M for each memory of each image data held in e3
/'+, MA2. MA3 is required. Then, the normalization circuit 17 quotients the image data in the memory e1 by MA+,
Image data D+, D2 . D3 is connected to memory Ml, M2 . M3
to hold. In this way, the light sources PI are stored in the memories M1 to M3.
-P Three sets of normalized image data obtained when each light is turned on are held.

The orientation determination circuit 19 is in the reference table creation mode.

Based on the contour data of the spherical object 13 transmitted from the center position/radius calculation circuit 14, the orientation (normal vector) of each surface element of the object 13 is sequentially read out from the inclination table 20 and stored in the memory M+-M3. A set of image data D1 to D3 read out sequentially for each pixel is added and stored in the reference table 21. For example, as shown in FIG. 2, the direction LI at pixel rl is read out from the tilt table 20, and the image data dl-1+d2-I at the same pixel r1 is read out from the memories M1 to M3.
, a3-1, and this (LI:d+-+.

dz-1+ a3-+) is stored in the reference table 21.

Similarly, the combination of orientation and image data obtained for images @rz and r3 (L2: d+-2, dz-2゜d:+-
2), (L3: d+-3, dz-3・d3-3
) is stored in the reference table 21. By performing this for all pixels in the contour data, the combination of image data D+ -D3 for the orientation of the two-plane elements is determined,
Based on this, the direction of the surface element can be detected from a combination of three image data. At this time, if the three components of the image data are )x, , 3y, and az, this can be expressed as (p, q). (
Two directions are generally expressed as -1. (2) Creation of surface orientation of object to be inspected After creating the reference table 21 according to (1) above, the object to be inspected 23 is placed at the position of the object 13 shown in FIG. Then, the image data obtained from the TV camera 12 by turning on the light source P+ is held in the memory m1, the image data obtained by turning on only the light source P2 is held in the memory m2, and finally the image data obtained by turning on the light source P3 is held in the memory m1. The image data obtained by turning on only the light is held in the memory m3. Then, as in (11) above, these image data are corrected by the correction circuit 15 using the correction coefficient table 16, and the corrected image data is held in the memories el, e2, and e3.
The maximum value search circuit 18 searches for the maximum value for each l −83,
Using the obtained maximum values, the normalization circuit 17 stores the memories el, e2 . After normalizing the image data held in e3, this normalized image data is stored in memory M.
s, M2. Store in M3. At this time, direction determination circuit 1
Since M1.9 is controlled in the surface element orientation detection mode, each memory M1. M2. Image data at the same address (pixel unit) is read out by sequentially scanning M3. At this time, for example, dl-# from memory M1, dz-n from M2, M
When image data d3-m is obtained from 3, these (d+-1, dz -n.

d3-m) as a set and accesses the reference table 21.

The orientation (normal vector) Lf of the surface elements corresponding to these image data sets is obtained as output data.

In this way, by comparing the set of pixel units in the memories M1 to M3 with the reference table 21, the orientation of each surface element can be obtained. FIG. 3 is an example of graphical output of the orientation of surface elements when a truncated cone is used as the object to be inspected.

In the present invention, the orientation of the surface elements of the entire screen is determined using the methods (1) and +2) above, and based on this, local defects (scratches) can be detected and global defects (gentle curves) can be detected. , distortion, etc.), one embodiment of the present invention will be described with reference to FIG.

In FIG. 4, the parts with the same symbols as those in FIG. 3 is a model holding section, and 33 is a comparison circuit.

The orientation determination circuit 30 performs the same processing as the orientation determination circuit 19, but has a local defect detection mode and a model creation mode in addition to the reference table creation mode.

Controlled by global defect detection mode, etc. In the case of the local defect detection mode, the orientation of each surface element of the object to be inspected is detected and outputted to the differentiator circuit 31, as in the case of the surface element orientation detection mode in the orientation determination circuit 19. In the model creation mode, the orientation of surface elements of a defect-free model necessary for global defect detection is detected and output to the model holding unit 32. In the case of the global defect detection mode, the orientation of each surface element of the object to be inspected is detected and sent to the comparison circuit 33, and the orientation of each surface element of the model is first stored in the model holding section 32. It is compared with.

The differentiator circuit 31 calculates the direction of the surface element on the surface of the object (normal vector (
p, q)) is differentiated, a dark value is set to the magnitude of the differential value, and areas larger than that are detected as flaws, and a known differential calculation circuit is used. As an example, if the surface element of the relationship shown in FIG.
(Multiply by a coefficient such as C1.

It performs calculations. Of course, the differential calculation circuit is not limited to the one exemplified here, but the differential value Sij can be obtained as follows.

5pxij = ((pi+++j-+ -pi-++
j−+)+2 (1)! +l+J G'l-1+
J ) + (pi++1j÷+ pi−+J+
+)) Spyij = ((pi-t・j÷s
pi-t+j-t)+2 (pLj++ pi
, j−+)+ (pi++Ij++ −pi+++j−
t) )Sqxij = ((qi+++j−+ −
qi-++j-+)” 2 (Q1+I+J Ql
−1+ J )+ (qL++j++ qi−+・
j++)) Sqyij = ((qi-++j++
qi-++j-+)+ 2 (qi, j++
qi+j-+)+ (qi+++j++ qi+
tlj−+))Sij= 5pxij2+5pyij
2+5qxij2+5qyij2−−−−−−−43)
If the value obtained from equation (3) is large, it indicates a large rate of change in the direction of the adjacent surface element, which indicates the presence of a flaw, so this Sij is determined by the darkness value as described above to identify the flaw. The presence or absence can be determined.

Next, the operation of the configuration of the embodiment of the present invention shown in FIG. 4, that is, the detection of local defects (scratches) and the detection of global defects (gentle bends, distortions, etc.) will be described.

10 Detection of local defects (scratches) For example, in FIG.
When detecting whether or not scratches exist.

In the same manner as in (1) above, the orientation determination circuit 30 is controlled in the reference table creation mode, and a reflector having a known shape, for example a spherical one, is placed below the TV camera 12 with specular reflection similar to the object to be inspected 34. Place an object 13 with characteristics and refer to the reference table 21.
Create. Then, object 13 is replaced by object 3 to be inspected.
4, the light sources PI to P3 are sequentially turned on individually, and the direction determination circuit 30 is controlled in the local defect detection mode.

Thereby, the orientation determination circuit 30 operates in the same manner as in (2) above, and sequentially sends the orientations of the surface elements obtained from the object to be inspected 34 to the differentiating circuit 31. The differentiating circuit 31 calculates a differential value based on the direction of the thus transmitted surface element based on the above-mentioned equation (3). For example, the direction determination circuit 30 may
) is obtained, and this is determined by the differentiator circuit 3.
1 as described above, a differential value as shown in FIG. 6 (bl) is obtained, and this is judged by a predetermined darkness value.

One local defect can be output as shown in FIG. 6(C). In this way, the presence or absence of local defects such as scratches can be detected by determining the differential intensity using equation (3) and determining the value using the dark value.

(2) Detection of global defects For example, when detecting global defects in the object to be inspected 35' in FIG.
Create 1. Then, the orientation determination circuit 30 is controlled in the model creation mode to
5 as a model, the orientation of the surface elements of the defect-free object 35, which is this model, is determined in the same manner as in (2) above, and this is stored in the model holding unit 32. A defect-free object is an object that is exactly as designed. Next, the orientation determination circuit 30 is controlled to the global defect detection mode, the object to be inspected 35' is placed in place of the defect-free object 35, and the image data is stored in the memories M1, M2, . Let M3 hold it. Direction determination circuit 30
The comparison circuit 33 compares the orientations of each isoplane element of the model defect-free object 35 and the object to be inspected 35'. In this case, since the orientation of the surface elements of the defect-free object 35 is held in the model holding unit 32, it is sufficient to read it. Inspection object 35
The direction of the plane element ' may be stored in a memory (not shown) prior to the comparison, or may be obtained by accessing the memories M1 to M3 and the reference table 21 each time the comparison is made.

In order to speed up processing, it is desirable to store the data in memory before comparison. In order to perform this comparison, it is necessary to align the defect-free object 35, which is a model, and the object to be inspected 35', but this can be done by aligning the centers of gravity. The following equation (4) was used as the equation for calculating the degree of mismatch in the comparison circuit 33. Here (po
, q o> is the normal vector of the object to be inspected 35', and (pm.

qm) is a normal vector of the defect-free object 35 that is a model.

dij −((poij−pmij) 2+ (qo
ij −qmij) 2+ (poij −pmi−1
, j−+) 2+ (qoij −qmi−+++j−+
) 2+ (poij −pmi, j−+) 2+ (
qoij −qmi, j−+) 2+ (poij
-pmi-4,j-1) 2+ (qoij -qmi
+1+j-+) 2+ (poij-pmi-t,
j) 2+ (qoij −qmi−++j) 2+
(poij −pmi+4.j) 2+ (qoij
−qmi++, j) 2+ (poij −pmi−
t+j++) 2+ (QOij Qmi−+
+jn) 2+ (pojj -pmi, j+
+) 2+ <qojj −qmi, j+1)
2+ (poij −pmi+4.j+1)
2+ (qoij −qmi++, j++) 2)
/9-1--・--(41 Then, this degree of mismatch is determined for the entire surface element using the following equation (5).

dshi Σd i j −−−−−
---151 A darkness value is set for this d, and thereby global defects can be detected. For example, when the model is a truncated cone, it is possible to detect even if there is a mild defect such as an object to be inspected 35' in FIG. 4 having a V-shaped side surface.

In this case, the model has the direction of the surface elements as shown in FIG. 7, and the object to be inspected 35' has the direction of the surface elements as shown in FIG. Figure 9 (
As shown in a), by slicing every surface element using the darkness value, it can be shown as shown in FIG. 9 (bl).

By the way, the computation in the comparator circuit 33 is performed by the above equation (4),
In addition to outputs based on equation (5) and their respective dark values, it is also possible to obtain outputs based on calculations of linear equations (6) and (7) and their respective dark values.

dpij= ((QOij pH1ij) + (p
oij −pmi−1,j−1)+(poij−ρm
i, j-1) -4-(poij -pmi+1+j-+)+ (po
ij −pmi4・j) + (poij −pmL++j) + (poij plTli−1+j+1)+
(poij pmi +j+4)+(POIJ
pmL+l+J+1) ) −'−'−'(6) dQ
ij= ((QOij qmij) + (qoij
-qmi-1,j-1)+ (qoij -qmi,
j−t)+ (qoij −qmi+1+j−+)
+ (qoij qmi-++j) ” (Qoij qmiiyaIIJ) + (qoij -qmi-++L+)+ (qoij
−qmi+ j++)+ (qoij −qmi+h
j++) >−−−−−−−−−(7>> By the way, the above equations (6) and (7) indicate in which direction the direction of the surface element is inclined.

FIG. 10 shows the equations (6) and (7) in which only positive or negative values are output.

〔Effect of the invention〕

According to the present invention, by using a diffused reflection surface and a plurality of light sources, it is possible to detect not only scratches on the surface of an object having specular reflection characteristics such as a mirror surface, but also defects such as gradual loosening and bending in an extremely accurate manner. be able to. Therefore, specular objects, which conventionally relied on visual inspection, can be accurately inspected using the data processing method.

[Brief explanation of drawings]

Figure 1 (a) shows an image data input device for inputting image data in the present invention (bl is a flowchart for determining the center position and radius of an object (C)). (d) is the direction of the surface element. Explanatory diagram when extracting, same (e)
Figure 2 is a block diagram for creating a reference table and outputting the orientation of surface elements, which is a feature of the present invention, and Figure 2 is a reference table creation form! :,
3 is an example of the extraction result of the direction of the surface element obtained from FIG. 1, FIG. 4 is a configuration diagram of an embodiment of the present invention, FIG. 5 is an explanatory diagram of a differential circuit, and FIG. Local defect detection example, 7th
The figure shows the orientation of the surface elements of the model, Figure 8 shows the orientation of the surface elements of the object to be inspected, Figure 9 shows an example of discrepancy detection, and Figure 10 (al,
(b) shows the deviation in the X-axis direction when viewed from above;
(d) shows the same bias in the y-axis direction, and FIG. 11 is an explanatory diagram of conventional photometric stereo. In the figure, 1 is an object, 2 is a TV camera, 10 is an image data input device, 11 is a diffused reflection surface, 12 is a TV camera, 13 is an object, 14 is a center position/radius calculation circuit, 15 is a correction circuit,
16 is a correction coefficient table, 17 is a normalization circuit, 18 is a maximum value search circuit, 19 is a direction determination circuit, 20 is a tilt table, 21 is a reference table, 22 is an output data section, 23 is an object to be inspected, 30
31 is a differentiation circuit; 32 is a model holding unit; 33 is a comparison circuit, 34 is an object to be inspected, 35 is a defect-free object (model), and 35' is an object to be inspected. Figure 2 Figure 6 ((:l) CC) Figure 7 Figure 8 (b) Figure 10

Claims (3)

[Claims] (1) At least three light sources, a diffuse reflection surface that irradiates a target object with light by reflecting light from the light sources, an image input means for observing the target object, and a known-shaped object having regular reflection characteristics. a reference table creating means for creating a reference table that expresses the relationship between the set of grayscale values observed under each light source point light and the orientation of the microscopic surface elements on the surface of the object; and a reference table that holds the created reference table. The direction of the micro-area element on the surface of the object to be inspected is extracted by searching the reference table holding means using the holding means and a set of grayscale values observed under each light source point light of the object to be inspected having specular reflection characteristics. 1. A method for detecting defects on the surface of an object having specular reflection characteristics, comprising: an extraction means; and a defect on the surface of the object to be inspected is detected based on the output of the extraction means. (2) The defect detection is performed by detecting local defects on the object surface by extracting discontinuities in the orientation of the micro-area elements extracted by the extraction means. A method for detecting defects on the surface of an object having specular reflection characteristics as described in (1). (3) The defect detection is performed by extracting the orientation of the micro-area elements of the model by the extraction means, and by extracting the orientation of the micro-area elements of the object to be inspected and comparing these. 2. A method for detecting defects on the surface of an object having specular reflection characteristics as claimed in claim (1), characterized in that the method detects defects such as: