JP2738252B2 - Edge detection device of strip material by image - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001] The present invention relates to an apparatus for detecting an edge of a strip by taking an image of an end (edge) of the running strip and detecting an edge from the image.

[0002]

2. Description of the Related Art Processes for handling steel sheets and strips,
For example, in pickling, galvanizing, etc., the ends of steel plates or strips (strips) are detected to take up the coil, send out the ears in a line, or prevent meandering during transport, and the strip Control is performed so that winding, unwinding, and conveyance are performed normally. Such controls deal with strips, paper industry, printing and binding industry, textile industry,
It is widely used in the plastic and film industries. In this case, it is necessary to accurately detect the edge of the strip.

As an edge detection method, an edge portion of a running strip is imaged by a television camera, and it is determined whether a boundary between light and dark at the edge portion is an edge based on a certain threshold value.

[0004]

As described above, when an edge portion is imaged by a television camera and an edge is detected from the image data, the edge cannot be often detected under the following conditions. The image is out of focus. Video input level is low. Lighting is uneven. The subject (strip) is dirty. The image contrast is poor. There are two or more locations that show a density difference that may be considered an edge. Light-to-dark or dark-to-bright in the direction of the edge density change is sometimes inverted.

An object of the present invention is to provide an apparatus for detecting an edge of a band-shaped material by using an image for detecting edges with high accuracy by performing various types of data processing even under the above-mentioned bad conditions.

[0006]

In order to achieve the above object, an A / D converter for converting an analog signal output from a television camera into a digital signal, and a method for displaying the converted data on a screen, A one-dimensional pixel array generating means for averaging the data of the pixels at the same position in the horizontal direction for the pixels on the number of horizontal scanning lines to obtain a one-dimensional pixel data array; and k pixels preceding the pixel of interest in the one-dimensional pixel data array ,
Taking k rear pixels, multiplying each by a predetermined coefficient, obtaining the sum of the previous pixel group and the rear pixel group, then obtaining a difference value pixel data sequence using the difference between the two as a new pixel of interest, The first difference pixel data sequence to the nth difference pixel data sequence according to the value of
A corrected pixel data sequence obtained by weighting each constituent pixel according to the difference between the position of each constituent pixel of each difference pixel data sequence on the horizontal scanning line and the already calculated edge position is a first corrected pixel data sequence Pixel data string generating means for generating the corrected pixel data string as the n-th corrected pixel data string, and a pixel position having the maximum value of each corrected pixel data string as an edge position candidate of the corrected pixel data string. Edge position selecting means for selecting one edge position based on a judgment criterion.

In addition, as the predetermined criterion, the relationship between the maximum value of each corrected pixel data string and the next largest value is examined.
An edge position candidate of a corrected pixel data string having a large difference between the two is selected.

As the predetermined criterion, each corrected image data string is represented by a curve, and the center position of two intersections that intersects a curve near the edge position candidate with a predetermined level threshold value is set as a new edge position candidate. Is selected.

In addition, the edge position candidate is selected in consideration of the short distance between the new edge position candidate and the already calculated edge position.

Further, as the predetermined criterion, an edge position candidate of the corrected pixel data string having the largest value among the maximum values of each corrected pixel data string is selected.

As the predetermined criterion, the degree of reliability is evaluated in consideration of calculation accuracy and noise for each corrected pixel data sequence, and when selecting an edge position candidate of the corrected pixel data sequence, The level of reliability is taken into consideration.

Further, as the predetermined criterion, for each edge position candidate of each corrected pixel data string, it is checked whether the direction of the density change of the edge position candidate is from light to dark or from dark to light. Is the same as the change in the density direction of the edge position already calculated, the same edge position candidate is selected as having high reliability.

[0013]

A one-dimensional pixel array generating means for converting an analog signal output from a television camera into a digital signal by an A / D converter, when displaying the converted digital on a screen,
For a predetermined number of continuous pixels, for example, 64 pixels on a horizontal scanning line, data of pixels at the same position in the horizontal direction is averaged in the vertical direction to obtain a one-dimensional pixel data string. In other words, taking pixels on 64 scanning lines as an example, each pixel on a horizontal scanning line creates a one-dimensional pixel data string composed of an average value of 64 pixels. Thereby, the influence of noise can be reduced. The pixel data string generating means is configured to output the pixel of interest of the one-dimensional pixel data string.
(Arbitrary pixel) Take the k pixels arranged before and after, and multiply each of the k pixels by a predetermined coefficient, and calculate the sum of the front pixel group and the sum of the rear pixel group. Then, a difference value pixel data string is determined using the difference between the two sums as new data of the target pixel. This difference value pixel data sequence emphasizes the density change of the pixels on the horizontal scanning line. A difference between the position of each constituent pixel on the horizontal scanning line and the already calculated edge position is obtained for the difference pixel data string, and weighting according to the difference is performed on each constituent pixel to create a corrected pixel data string. Usually, the edge position does not largely change. If the edge position greatly changes, it is often caused by noise. With respect to each of the corrected pixel data strings that differ depending on the value of k obtained in this way, the pixel position having the maximum value for each of the corrected pixel data strings is set as an edge position candidate of the corrected pixel data string. Based on this edge position candidate, one edge position candidate is selected as a correct edge position by the edge position selection means according to a predetermined criterion.

As a predetermined criterion, the relationship between the maximum value and the next largest value is checked for each corrected pixel data string, and an edge position candidate of the corrected pixel data string having a large difference between the two is selected. The edge position candidate is often the position of the pixel indicating the maximum value, but if there is another value close to this maximum value, it is highly likely that it has been caused by noise. Therefore, as the difference between the maximum value and the next largest value is larger, the data has less noise. Therefore, an edge position candidate determined by data having less noise is selected as a correct edge position candidate. The noise here is electrical, uneven lighting or dust,
These include flaws, changes in the surface state of the inspection target, and changes in the shape of the edge.

As a predetermined criterion, when each corrected image data string is represented by a curve, even if there is disturbance such as noise, the edge position candidate is determined by a curve representing the vicinity of the edge position candidate and a threshold of a predetermined level. It becomes closer to the center of the two intersections that intersect. Therefore, when selecting an edge position candidate, this center point is selected as a new edge position candidate.

Further, by selecting an edge position candidate in consideration of whether or not this new edge position candidate is close to the already calculated edge position, a more reliable selection can be made.

As a predetermined criterion, a candidate edge position of the corrected pixel data string having the largest value among the maximum values of the corrected pixel data strings is selected. Since the edge position is the position having the largest difference in shading, the maximum value of each corrected pixel data string is compared, and the edge position candidate of the corrected pixel data string having the largest value is determined as the correct edge position.

Regarding the difference pixel data string, it is described that the smaller the value of k, that is, when k = 1, the most emphasized is the difference in shading of pixels, and it has been described that the greater k, the smaller the degree of emphasis. However, this also applies to the corrected pixel data string as it is. On the other hand, the influence of noise is larger as k is smaller, and is smaller as k is larger. As described above, from the calculation accuracy, the edge position candidate determined from the corrected pixel data string of k = 1 is good, but the accuracy is reduced when noise is considered. This noise depends largely on the shooting environment of the TV camera, such as lighting, as well as changes in the shape of the edges, dust, and scratches. Therefore, in consideration of calculation accuracy and noise, the level of reliability of the edge position candidate of each corrected pixel data string is determined, and the edge position candidate is selected.

The edge position is detected by photographing the running strip with a television camera and sampling the data at predetermined intervals. For this reason, there is little possibility that the direction of the change in the density difference, that is, from bright to dark or from dark to bright, will correspond to the one-dimensional pixel data string between the previously detected edge position and the currently calculated edge position. Therefore, an edge position candidate of the corrected pixel data string in which the direction of the change of the density difference does not change is selected as having high reliability.

[0020]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an image processing apparatus for realizing an embodiment of the present invention. A television camera 1, an A / D converter 2 for converting input data from the television camera 1 from analog to digital, an input buffer 3 for storing the converted data, a bus 4, and a CPU 5 for controlling the whole
An image processor 6 for processing image data;
A program memory 7 for storing a program for determining the operation of the PU 5, a grayscale data memory 8 for storing data subjected to grayscale processing by the image processor 6, an output buffer 9 for temporarily storing output data, and digital data of the output data. D / A converter 10 for converting to analog data
And a CRT 11 for displaying output data.

FIG. 2 to FIG. 5 are diagrams showing the flow of the processing of this embodiment. Hereinafter, description will be made with reference to these figures. In FIG. 2, an edge portion of a running strip taken by an ITV (industrial television camera) 1 is converted from analog data into digital data by an A / D converter 2.

FIG. 6 shows a method of sampling data used for analysis after image data. As shown in FIG. 6A, data on 64 scanning lines near the center in the vertical direction in one screen scan is employed. . If it takes, for example, 1/6 second to detect one edge by processing this data, the next data is adopted as shown in (b) 1/6 second after the screen data shown in (a) is adopted. Data on the center 64 scanning lines of one screen scan is adopted. Note that 64 is an example, and a different number may be used. Data for each scanning line is converted into, for example, an 8-bit binary signal.

In the integrating stage 20 shown in FIG. 2, these 64 data are averaged. In this method, pixels at the same position on a scanning line are added 64 times. One line of data is stored in a one-line memory, and is added to the next input one-line data. Line data is added. By performing the addition in this manner, a one-dimensional data string in which data is averaged and random noise is reduced is obtained.

The level calculation stage 21 in FIG. 2 is a process for extracting the maximum amplitude of data. The edge position is where the difference in shading between adjacent pixels is large. Therefore, since the difference between each pixel becomes a problem, the maximum value of the data as the maximum amplitude,
The minimum value is taken, and the variation of data between the maximum amplitudes is examined.

FIG. 7 is a diagram showing the calculation of the level (maximum amplitude) of a one-dimensional data string. The horizontal axis represents the position of a pixel on a scanning line,
The vertical axis indicates the gray level indicating the density of the pixel. The difference between the maximum value and the minimum value is defined as the maximum amplitude, and the offset is removed by subtracting the minimum value from each data.

When the maximum amplitude thus obtained is small, noise becomes relatively large. Therefore, averaging is also performed in the horizontal direction to further remove noise depending on the magnitude of the maximum amplitude. This is to add the target pixel (target pixel) and its neighboring pixels to average the target data and improve the S / N ratio.

The conditions for performing the averaging process and an example of the process will be described. Condition 1: maximum amplitude ≧ maximum value of data / 4 no processing Condition 2: maximum value of data / 4> maximum amplitude ≧ maximum value of data / (8√2) Smoothing 3 is performed by equation (1). Condition 3: maximum value of data / (8√2)> maximum amplitude ≧ maximum value of data / 32 Smoothing 5 is performed by equation (2). Condition 4: Maximum amplitude <maximum value of data / 32 Since an edge cannot be detected, an error occurs.

In the smoothing process, the following equation (1) and (2) are used for D0 at the center of the data arranged as shown in FIG.
This is done by an expression. (D-1) + D0 + (D + 1) (1) (D-2) + (D-1) + D0 + (D + 1) + (D + 2) (2)

The data on which the smoothing 3 and the smoothing 5 have been performed by the equations (1) and (2) are used to calculate the level described with reference to FIG. 7 again.

Next, referring to FIG. 3, the data obtained in the level calculation stage 21 is expanded in a data expansion stage 22. This decompression process processes data in accordance with the capability of the computer. For example, if the computer has a processing capability of 16 bits, if the data obtained in the level calculation stage 21 is 14 bits, this data is
This is the process of expanding to bits.

In FIG. 3, the expanded data is differentiated by a differentiation filter stage 23 to calculate a difference between adjacent pixels. A plurality of filters are provided as the differential filters. In this embodiment, five types are shown, but it is not necessary to limit to this number.

Equation (3) shows the product-sum operation in the range 3.
Equation (4) shows the product-sum operation in the range 5. ED3 (*) = D (-1) -D (1) (3) ED5 (*) = (5D (-2) + 3D (-1) -3D (1) -5D (2)) / 8 ( 4) Here, it is assumed that the data whose levels have been calculated are arranged as follows. D (-2), D (-1), D (*), D
(1), D (2)

The product-sum operation in the range 9 is shown in equation (5). ED9 (*) = (13D (-4) + 10D (-3) + 6D (-2) + 3D (-1) -3D (1) -6D (2) -10D (3) -13D (4)) / 32 ... (5)

The product-sum operation in the range 17 is shown in equation (6). ED17 (*) = (6D (-8) + 6D (-7) + 5D (-6) + 5D (-5) + 4D (-4) + 3D (-3) + 2D (-2) + D (-1) -D (1 ) -2D (2) -3D (3) -4D (4) -5D (5) -5D (6) -6D (7) -6D (8)) / 32 ... (6)

In the range 17R, the data is divided by 16, and the data is smoothed in the same manner as shown by the equations (1) and (2).
This is a process in which step 15 is performed to further include a wide range of data, and the product-sum operation process in the range 17 is performed on the data.
This is ED17R.

This differential filter process is basically a process for extracting the amount of change in data. If the amount of change is simply regarded as “large change rate”, the change due to noise may be recognized as being larger than the original edge, so that the target pixel and its neighboring pixels are replaced with a simple high-pass filter. A band pass filter based on a product-sum operation between the two is used. This calculation is performed between several types of pixel ranges, and the feature of each calculation range is extracted.

9 to 11 show data of the product-sum operation processing in the range 3 to 17R. In FIGS. 9 to 11, the uppermost row shows a level calculated in the level calculation stage 21 and then data expanded in the data expansion stage 22. A indicates range 3, B indicates range 5, C indicates range 9, D indicates range 17, and E indicates range 17R. FIG. 9 shows a case where data that can be lightened from the left is darkened at the edge, and FIG. 10 shows a case where data that has been darkened from the left is lightened at the edge. FIG. 11 shows a case in which A and B have two maximum values, so that an edge cannot be determined in A and B and the edge is C or later. 9 and 10 show examples in which an edge is determined after A.

The position weighting stage 24 in FIG.
This is a process of weighting each data subjected to the processing of the range 17 to 17 according to the distance from the edge. The edge positions used here are the edge positions obtained in the previous measurement described with reference to FIG. A template corresponding to the small deviation from the edge position obtained last time is created, a product of the template data and the raw data is created, and the "edge-likeness" of a portion close to the previous edge is emphasized. . This processing is based on the premise that the position of the edge to be detected this time from the previously detected edge position usually does not change much. Note that the previous edge position may be set to a value obtained by averaging previous and previous edge positions or an average of several previous edge positions.

FIG. 12 is a diagram showing an example of a weighting curve when a position template is created. (A) is an example using a SIN curve as a curve, and (b) is an example using a polygonal line. The horizontal axis indicates pixels, which are 0 to 1000. This means that there are almost 500 pixels obtained in the differential filter stage 23, the top point of the curve is TOP, and the position of this TOP is the position of the edge so that it is possible to cope with the case where the edge comes to the left or right end. That is, the position of the pixel serving as TOP is set as the edge position obtained in the previous processing, and each pixel is weighted to this position according to the SIN curve or the polygonal line. Therefore, the pixel at the position of the TOP is given the maximum weight, and as the distance from the pixel increases, the value of the weight changes according to the SIN curve or the polygonal line. The weighting curve may be a SIN curve or a polygonal line depending on the state of the edge to be detected, but may be another curve. The data up to this point is used in a determination process at a later stage, and is stored in a memory or the like and used as a template. This stored data is wda
Let it be t.

The reason why the weighting process is performed is based on the fact that the next edge is usually in the vicinity of the position determined as the edge in the previous detection, and there is little sudden change as described above. A sudden change may occur when a splice or a notch mark is made. Conversely, when a change point like an edge is found at a position distant from the edge, there is a high possibility that the change point is caused by noise such as a flaw or uneven illumination, so this processing is to cancel this.

Next, the candidate point calculation stage 25 of FIG. 4 will be described. In the position weighting stage 24, data of five regions ranging from 3 to 17R can be obtained. Since each range k has a positive region and a negative region, edge position candidates in each region are detected for a total of ten regions. In detecting these edge position candidates, level calculation and data expansion are performed for the ten regions obtained in the position weighting stage 24. These are the same as the processes performed in the level calculation stage 21 and the data decompression stage 22.

Next, detection of a candidate edge position for each area will be described. As a detection method, one edge position is determined from a plurality of candidate points using a scoring method. That is, each candidate is evaluated from a plurality of viewpoints, and the one with the highest overall score is determined as a true edge. In each area, the edge position candidate is the position of the data at which the maximum value of the level calculation and data expansion is obtained. However, when there is noise due to variations in the shape of the edge or a change in illumination, a value close to the maximum value is generated in an area other than the edge portion, and it is difficult to determine the edge position. Therefore, the position of the data having the maximum value is determined as the edge position candidate.If there is no other value close to the maximum value to indicate the probability, a large score is given assuming that the edge position candidate has high reliability, and the maximum value If there is another value close to, we will give a small score as unreliable. As a result, among the edge position candidates output from the ten regions, the edge position candidate with a high score can be determined to be the most reliable edge position.

An example of a method for giving this score will be described below. FIG. 13 is a diagram illustrating a method of calculating a score for a candidate edge position in each region. (A) shows the case where there is no other value close to the maximum value, and (b) shows the case where there is another value close to the maximum value. a is the maximum value of the data. This is set as an initial value. c is a reference value (for example, maximum value / 2). Based on this, it is compared whether each data exceeds the reference value.
If there is no data exceeding the reference value other than the maximum value, a predetermined score is given. As shown in (b), when there is a point b next to a between c and a, a score inversely proportional to the height of b (a value smaller than the predetermined score) is attached. In addition, negative data is scored by judging the magnitude based on its absolute value. In this manner, edge position candidates and their scores are obtained for each of the ten regions. This score is indicated as the score.

Next, the score calculation stage 26 of FIG. 4 will be described. This score calculation stage 26 has the following scores,,
Is calculated. One of them is position weighting stage
When the data of range 3 to range 17R obtained in 24 is represented by a curve, a point that generates the maximum value that is an edge is an edge candidate, but when there is disturbance such as noise, the position of the maximum value is not an edge There is also. FIG. 14 shows an example of such a case, in which (a) shows the data obtained in the level calculation stage 21 of FIG. 2 and (b) shows the data obtained in the position weighting stage 24. The portion indicated by A in (a) is caused by disturbance, and the portion where such a rapid change occurs has a maximum value as shown in (b).

For this reason, there may be cases where the maximum value cannot be simply used as an edge position candidate. This is because the shape of the edge of the plate is not constant, and the position of the maximum value changes depending on how the illumination light is illuminated. For this reason, two points that intersect the threshold value of the predetermined level are obtained for the curve shown in FIG. 14B, and the center position of the two points is set as a new edge position candidate. FIG. 15 is a diagram illustrating this situation. As the predetermined level, a good result can be obtained by empirically using the larger value of the following. 7/8 × initial value (initial value + search value) / 2 Here, the initial value is the maximum value of the data and is the value of a in FIG. 13, and the search value is the value of b in FIG.

For the new edge position candidate obtained in this way, the distance from the previously calculated edge position (or the average of the edge positions calculated several times before) is calculated. It is determined that the accuracy of the position is high.

As an example of a method for performing the above-described empirical reliability evaluation of the edge position candidates by using a score, an example of a new edge position candidate and a method of calculating a score in consideration of the difference between the new edge position candidate and the previously calculated edge position Will be described. FIG. 15 is an explanatory diagram for calculating a new edge position candidate from a curve that produces a maximum value. The maximum value of the data is an initial value a, and the search value is b. Here, the search value is the value of c in FIG. 13A and the value of b in FIG. The edge position d is obtained from a and b as follows, and the score is calculated. In FIG. 15, d 'is the position of the maximum value, and d is a new edge position candidate.

(1) The larger one of 7/8 × initial value and (initial value + retrieval value) / 2 is set as the value for obtaining the edge position. (2) The intersection of this value and the curve representing the edge is c, c1, and (c + c1) / 2 is the edge position. (3) The absolute value of the difference between this edge position and the edge position obtained in the previous processing is taken and used as gap data G.
However, when G> 100, G = 100. (4) The score is calculated by the following equation. Score = (100−G) / 10 This process is also performed for negative data. In addition, score ≧
Set to 0.

Next, the wdat of the data stored in the memory is called in the position weighting stage 24 of FIG. 3, the maximum value of each area is obtained, and a score corresponding to the maximum value is obtained, and this is set as the score.

Next, for the range 3 to range 17R obtained by the product-sum operation, a score is provided for the degree of reliability of the edge position candidate for each range. The product-sum operation is from equation (3)
As shown in the equation (6), the smaller the value of the range k, the more the difference in density between adjacent pixels is emphasized. Therefore, it seems that the edge position candidate obtained in the range 3 is most likely. However, the influence of noise is largest in range 3 and smallest in range 17R. Taking these both into consideration, a score is given in the range 3 to the range 17R.
Score this. An example of this score is shown with reference to FIGS. According to this, A (range 3) is 4, B (range 5) is 6, C (range 9) is 4, D (range 17) is 2, E
(Range 17R) is 0, and Range 5 is the degree of emphasis of shading difference,
The maximum score is given in consideration of the influence of noise. However, this score may also vary depending on the subject, the shooting situation, and the like, and must be determined according to the situation.

Next, a score is given because the degree of reliability is high or low depending on whether or not the direction of change of the density difference obtained in the previous processing matches the direction of change of the current density difference. FIG. 9 and FIG.
As shown in FIG. 10, as to whether the direction of the change in the gray level is from light to dark or from dark to light, the direction of the change in the gray level in the previous processing and the direction of the change in the gray level in the current processing are checked. The difference between the score for the match and the score for the match is obtained.
This score is taken as the score. Since it is normal that the direction of the change in the shading difference is the same in the previous processing and the current processing, this score is increased, and when they do not match, the score is determined as abnormal and the score is set to 0, for example.

In FIG. 5, the judgment stage 27 includes a score of ten edge position candidates output from each area in the candidate point calculation stage 25, a score obtained from wdat stored in the memory in the position weighting stage 24, and a bonus calculation. stage
One edge position is determined in consideration of the score obtained in 26. These scores have different properties, but in determining the value of the score, determine the value in consideration of consistency with other score values, and calculate the total score by adding the score ~. The edge position candidate in the region having the largest total score can be set as the edge position of the current processing. The thus obtained edge position of the current processing may be used as the true edge position, or the average position of the current processing edge position and the previous processing edge position may be determined as the true edge position. The averaging reduces the variation in edge position.

[0053]

As is apparent from the above description, the present invention integrates data on a plurality of scanning lines to obtain average one-dimensional data, and converts the average one-dimensional data into a plurality of differential filters to determine an edge candidate for each differential filter. Then, a score is given in consideration of the calculation method of each edge candidate and noise, and the edge is determined based on the score. Therefore, the edge can be determined with high accuracy.

[Brief description of the drawings]

FIG. 1 is a block diagram of hardware for realizing the present invention.

FIG. 2 is a diagram showing a process from an image input by a television camera to a level calculation stage.

FIG. 3 is a diagram showing processing from a data decompression stage to a position weighting stage.

FIG. 4 is a diagram showing processing in a candidate point calculation stage and a score calculation stage.

FIG. 5 is a diagram illustrating processing in a determination stage.

FIG. 6 is a diagram showing scan lines for collecting data.

FIG. 7 is an explanatory diagram for calculating a data level.

FIG. 8 is a diagram illustrating an arrangement of pixels.

FIG. 9 is a diagram illustrating an example of input data and a product-sum operation.

FIG. 10 is a diagram illustrating another example of input data and a product-sum operation.

FIG. 11 is a diagram illustrating still another example of the input data and the product-sum operation.

FIG. 12 is a diagram illustrating position weighting.

FIG. 13 is a diagram illustrating a process of narrowing down edge position candidates.

FIG. 14 is a diagram illustrating an example in which a position of a maximum value and an edge position candidate are different.

FIG. 15 is a diagram illustrating a method of calculating a new edge position candidate from a curve that produces a maximum value.

Claims (7)

(57) [Claims] An A / D converter for converting an analog signal output from a television camera into a digital signal, and for displaying the converted data on a screen, the same number of pixels on a predetermined number of continuous horizontal scanning lines in the horizontal direction are used. A one-dimensional pixel row generating means for averaging the data of the pixels at the position to obtain a one-dimensional pixel data row; and taking k pixels before and k pixels after the target pixel of the one-dimensional pixel data row, and , A sum of the preceding pixel group and the succeeding pixel group is obtained, and a difference pixel data sequence in which the difference between the two pixels is set as a new pixel of interest is obtained. The first difference pixel pixel is determined according to the value of k. From the data sequence to the n-th difference value pixel data sequence, weighting is performed on each component pixel according to the difference between the position on the horizontal scanning line of each component pixel of each difference pixel data sequence and the already calculated edge position. The corrected pixel data sequence to the first corrected pixel data sequence Pixel data string generating means for generating the corrected pixel data string as the n-th corrected pixel data string, and a pixel position having the maximum value of each corrected pixel data string as an edge position candidate of the corrected pixel data string. And an edge position selecting means for selecting one edge position based on a criterion. 2. As the predetermined criterion, the relationship between the maximum value of each corrected pixel data string and the next largest value is checked, and an edge position candidate of the corrected pixel data string having a large difference between the two is selected. 2. An apparatus for detecting an edge of a band-shaped material based on an image according to claim 1. 3. As the predetermined criterion, each corrected image data sequence is represented by a curve, and a center position of two intersections that intersects a curve near an edge position candidate with a predetermined level threshold value is set as a new edge position. 2. The apparatus according to claim 1, wherein the edge is selected as a candidate. 4. The strip-shaped material according to claim 3, wherein an edge position candidate is selected in consideration of a short distance between the new edge position candidate and an already calculated edge position. Edge detection device. 5. An edge position candidate of a corrected pixel data string having the largest value among the maximum values of each corrected pixel data string as the predetermined judgment criterion. For detecting the edge of a strip by using the image of FIG. 6. As the predetermined criterion, the degree of reliability is evaluated in consideration of calculation accuracy and noise for each corrected pixel data string, and when selecting an edge position candidate of the corrected pixel data string, 2. The apparatus according to claim 1, wherein the degree of reliability is considered. 7. As the predetermined criterion, for each of the edge position candidates of each of the corrected pixel data strings, it is checked whether the direction of the density difference of the edge position candidates is from bright to dark or dark to bright. 2. The image according to claim 1, wherein when the change direction of the edge position is the same as the change in the density difference direction of the edge position already calculated, the same edge position candidate is selected as having high reliability. Edge detection device for strips.