TWI689723B - Method for extracting dent on surface of object - Google Patents

Abstract

A method for extracting a dent on a surface of an object is disclosed and includes steps of: scanning an object; generating an image based on a scan result; dividing the image into a plurality of blocks in two dimensional directions; setting each of the blocks as a target block adjacent to a plurality of peripheral blocks, and calculating inter-block similarities between the target block and each of the peripheral blocks; generating a plurality of inter-block differences according to all inter-block similarities of each target block and a difference transformation function, and calculating a dent coefficient of each target block based on the inter-block differences and a number of the peripheral blocks; and extracting at least one target block as at least one dent block based on the dent coefficient of each target block.

Description

Method for extracting gravure on surface of object

The invention relates to an image processing method, in particular to a method for extracting gravure on the surface of an object.

In the production process of objects, if it is inconvenient to use labels as identification objects, gravure (such as printing or patterns, etc.) is usually left on the surface of the objects under high pressure or high temperature to facilitate identification of different objects. In addition, due to the low efficiency of manual recognition and poor error rate, an automatic recognition system has been gradually developed.

Taking the gravure characters on the surface of the steel embryo as an example, the conventional three-dimensional surface scanning method of the steel embryo scans the undulating image of the cross section of the steel embryo with a triangular laser optical measurement method, and uses the information of the undulation to identify the gravure area on the surface of the steel embryo. The known method is to generate a global statistical histogram from the measured data, use the histogram to find the numerical aggregation area where the gravure is located, infer the distribution range of the detected value where the gravure is located, and extract the pixel area that belongs to this numerical range in the image For analysis.

However, because the statistical histogram is not conducive to parallel operations, the above-mentioned conventional methods are not only time-consuming and lack of reliability; in addition, because the natural cross-section of the steel embryo has different surface uneven distribution, if the section is inclined, use the global histogram It will make the gravure area and the surface value of the whole area mixed, which is not conducive to highlight the numerical clustering of the gravure area, resulting in the inability to effectively extract the gravure character area.

In view of this, it is necessary to solve the problems of the conventional technology.

One object of the present invention is to provide a method for extracting gravure on the surface of an object, which uses the differences and correlations between local region features to find possible locations of the gravure, so as to remove the influence of global changes on the regional features, and thereby improve the gravure Effectiveness of extraction.

To achieve the above objective, an embodiment of the present invention provides a method for extracting gravure on the surface of an object. The method may be performed by an electronic system, which may include a control device, A storage device and a scanning device, the control device is electrically connected to the storage device and the scanning device, the storage device stores at least one instruction and data required to execute the process and the generated data to enable the control device to perform actions, The method may include: a scanning step, which causes the control device to cause the scanning device to scan an object; an imaging step, which generates an image based on a scanning result by the control device or the scanning device; and a partition step, which is controlled by The device divides the image into two blocks in two dimensions; an analysis step is performed by the control device for each block: the block is set as a target block, and the target block is adjacent to several peripheral areas Block, calculate the similarity between the target block and each peripheral block; a conversion step is performed by the control device for each target block: the similarity between all blocks of each target block Bring in a difference conversion function to generate the difference between several blocks of each target block, generate a total value according to the difference between the blocks, and calculate the total value according to a number of the peripheral blocks The average value serves as a gravure coefficient for each target block; and a printing step, the control device extracts at least one target block as at least one gravure block according to the gravure coefficient of each target block.

In some embodiments of the present invention, in the conversion step, the control device generates the total value according to the product of the difference between the blocks and the probability of an area of the target block.

In some embodiments of the present invention, in the conversion step, the printing area probability is calculated by a support vector machine algorithm based on the known gravure areas and known non-gravure areas in the multiple images.

In some embodiments of the present invention, in the conversion step, the printing area probability is based on a number of known gravure areas and a number of known non-gravure areas in a plurality of images using a K-nearest neighbor algorithm and a Gaussian Calculated by the model.

In some embodiments of the present invention, in the conversion step, the printing area probability is based on a number of known gravure areas and a number of known non-gravure areas in a plurality of images with a convolutional deep learning network algorithm Calculated.

In some embodiments of the present invention, in the analysis step and the conversion step, the control device calculates the inter-block similarity and inter-block difference of each target block according to the order of the target block.

In some embodiments of the present invention, in the analysis step and the conversion step, the control device calculates the inter-block similarity and inter-block difference of each target block in parallel.

In some embodiments of the present invention, in the partitioning step, the control device gives a two-dimensional coordinate to each block; and in the printing step, the control device is based on the gravure coefficient of each target block and the The two-dimensional coordinates extract at least one target block as at least one gravure block, and generate at least one mark in the image.

Through the above steps, the present invention can automatically find the parts with high probability of gravure on the surface of the blank (such as steel embryo) to form the gravure area, so as to further obtain the text or pattern content of the gravure area by using text recognition technology The present invention can also perform parallel operations, which has the effects of good accuracy and high processing efficiency. Compared with the conventional methods, it can improve the efficiency of blank identification and reduce the error of human eye identification.

Block B‧‧‧

B1‧‧‧ target block

B2‧‧‧Outer block

S1‧‧‧Scanning steps

S2‧‧‧Imaging steps

S3‧‧‧Partition steps

S4‧‧‧Analysis steps

S5‧‧‧Conversion steps

S6‧‧‧Printing steps

Figure 1: Schematic diagram of a method for extracting gravure on the surface of an object according to an embodiment of the invention.

Figure 2: Block diagram of an embodiment of the invention.

FIG. 3 is a schematic diagram of a target block and a peripheral block according to an embodiment of the invention.

Figure 4a: Experimental process image (1) of an embodiment of the present invention.

Figure 4b: Experimental process image (2) of an embodiment of the present invention.

Figure 4c: Experimental process image (3) of an embodiment of the present invention.

Fig. 4d: Experimental process image (4) of an embodiment of the present invention.

Figure 4e: Experimental process image (5) of an embodiment of the present invention.

Figure 4f: Experimental process image (6) of an embodiment of the present invention.

In order to make the above and other objects, features, and advantages of the present invention more comprehensible, the preferred embodiments of the present invention will be specifically described below in conjunction with the accompanying drawings, which will be described in detail below. Furthermore, the terms of direction mentioned in the present invention, such as up, down, top, bottom, front, back, left, right, inner, outer, side, surrounding, center, horizontal, horizontal, vertical, longitudinal, axial, The radial direction, the uppermost layer or the lowermost layer, etc., are only the directions referring to the attached drawings. Therefore, the directional terminology is used to illustrate and understand the present invention, not to limit the present invention.

Referring to FIG. 1, the method for extracting gravure on the surface of an object according to an embodiment of the present invention may include a scanning step S1, an imaging step S2, a partitioning step S3, and an analysis step S4, a conversion step S5 and a printing step S6.

Specifically, the method embodiment can be executed by an electronic system. For example, the electronic system can include a control device, a storage device, and a scanning device. The control device can be a device with data processing functions, such as a microcontroller or a microcomputer. Processor or special application integrated circuit (ASIC), etc.; the storage device can be a device with data storage function, such as memory or database; the scanning device can be a device with image capture function, such as: triangular laser An optical scanner or a three-dimensional scanning device in the form of structured light, etc., the control device is electrically connected to the storage device and the scanning device, and the storage device stores at least one instruction and data necessary for the execution process of the control device The information generated. The operation modes of the control device, the storage device, and the scanning device are understood by those with ordinary knowledge in the technical field to which the present invention belongs, and will not be repeated here.

As shown in FIG. 1, the scanning step S1 can be caused by the control device to cause the scanning device to scan an object (for example, a blank material such as a steel embryo or a mold embryo). The following only uses the steel embryo as an example, but not limited to this; for example, the scanning device can be used to perform multi-dimensional scanning on the object, and the result is transmitted to the control device. In this way, the scanning device can obtain data with the surface characteristics of the object.

As shown in FIG. 1, the imaging step S2 can be used by the control device or the scanning device to generate an image according to a scanning result. For example, the control device can arrange the two-dimensional data after scanning to generate the image, but Not limited to this, the image can also be generated by the scanning device and sent to the control device. In this way, a surface image of the object can be obtained based on the data of the surface characteristics of the object, which can be used as a basis for subsequent extraction of a gravure (such as printing or patterns on the steel embryo) on the object.

As shown in FIGS. 1 and 2, in the partitioning step S3, the control device can divide the image into two blocks B along the two-dimensional direction. For example, the block B is a block including many pixels. In this way, the entire image system can be divided into multiple local features to facilitate subsequent analysis. Among them, the control device can give each block B a two-dimensional coordinate, such as: (x, y).

As shown in FIGS. 1 and 3, the analysis step S4 can be performed by the control device for each block B: the block B is set as a target block B1, and each target block B1 is adjacent to several peripheral areas Block B2, calculate the inter-block similarity Sim ( x,y ) between the target block B1 and each peripheral block B2, the inter-block similarity can indicate the degree of correlation between two adjacent blocks, for example : The inter-block similarity (correlation) between blocks without gravure is high.

For example, as shown in FIG. 3, the peripheral block B2 is around the target block B1, for example: the block B connected to any edge or corner of the target block B1, usually each target area The block B1 is adjacent to the peripheral block B2 in 8 directions, but the target block B1 located at the edge of the image is less than eight adjacent to the peripheral block B2, for example: 3 or 5; the similarity between the blocks Sim ( x, y ) can be calculated as a cross correlation function or a spatial distance function (for example: |xy|/sqrt(|xy|)|, x and y are two-dimensional coordinates), the calculation method is the technology Those with ordinary knowledge in the field can understand, so I won't go into details here. .

As shown in FIG. 1, in some embodiments, in the analysis step S4, the control device may sequentially or in parallel calculate the similarity between the blocks of each target block.

其中，P(x,y)為二維座標(x,y)的目標區塊B1的凹印係數。 As shown in Figures 1 and 3, the conversion step S5 can be performed by the control device for each target block B1: the similarity Sim ( x,y ) between all blocks of each target block B1 is brought into a The difference conversion function G (.) is used to generate the inter-block difference degree G ( Sim ( x,y )) of each target block B1 according to the inter-block difference degree G ( Sim ( x,y )) A total value Σ G ( Sim ( x,y )), calculate the average value based on a number of the peripheral blocks (for example: N=8), as a concave of each target block B1 The coefficient is printed as follows:

Where, P(x, y) is the gravure coefficient of the target block B1 with two-dimensional coordinates (x, y).

For example, in the conversion step S5, the difference conversion function G (.) can be calculated using an inverse conversion or a normalized algorithm, for example: an inverse function or an inverse sigmoid function can be used to target The inter-block similarity between the block B1 and the surrounding blocks is converted into the inter-block difference between the blocks, and the inter-block difference is positively correlated with the degree of gravure between the blocks. In detail, since in an image, the characteristics of gravure between blocks are the opposite of those without gravure, for example, the similarity between blocks without gravure is high (ie, the difference is low), such as between blocks If the difference between blocks is low, the probability of gravure printing is low. On the contrary, if the difference between blocks is high, the possibility of gravure printing is high. Therefore, the lower the value of the difference conversion function G (.) calculation result (that is, the degree of difference between blocks) represents the higher the degree of similarity between blocks (the degree of correlation between blocks) (that is, the more related between blocks), For example: the position is also a non-gravure block; on the other hand, the higher the value of the difference conversion function G (.) (that is, the degree of difference between blocks) represents the similarity between blocks (the degree of correlation between blocks) Low (that is, the more unrelated the blocks are), for example: gravure blocks may appear.

其中，該印區機率V(x,y)為收集具有凹印的大量影像後，對該些影像中的凹印位置進行統計，而產生凹印在該些影像中的不同位置發生的機率，可用於進一步提升後續成功找到凹印的準確度；另，由於可能出現凹印區塊的區塊間特性為該塊間差異度高(即塊間相似度低)，因此，可僅採用該塊間差異度高(即塊間相似度低)處的區塊進一步進行上述印區機率的乘積計算，無須將所有區塊與印區機率進行計算，以利有效減少數據運算量。以下舉例說明該印區機率的多個實施態樣，惟不以此為限， 應被理解的是，該印區機率可依據多個影像中的已知凹印區域與已知非凹印區域以一支撐向量機(Support Vector Machine)演算法計算所得，例如：可事先收集多個具有凹印字區的影像，如：訓練一支撐向量機，可以得到一分類器，用以計算其屬於其中一類的比率。 In an embodiment, the control device can further be based on the difference G ( Sim ( x,y )) between the blocks of the target block B1 and the probability V ( x,y ) of a printing area of the target block B1 The product produces the total value Σ G ( Sim ( x,y ))× V ( x,y ), according to which, the gravure coefficient will correspond to the following:

Among them, the printing area probability V ( x, y ) is the probability of occurrence of gravure printing at different positions in the images after collecting a large number of images with gravure printing and counting the gravure printing positions in the images, It can be used to further improve the accuracy of subsequent successful finding of gravure; in addition, since the inter-block characteristics of the gravure block may be that the difference between the blocks is high (that is, the similarity between blocks is low), therefore, only the block can be used The blocks at high inter-block differences (that is, low inter-block similarity) are further subjected to the above-mentioned calculation of the product probability of the printing area, and it is not necessary to calculate the probability of all the blocks and the printing area, so as to effectively reduce the amount of data calculation. The following examples illustrate various implementations of the printing area probability, but not limited to this. It should be understood that the printing area probability can be based on known gravure areas and known non-gravure areas in multiple images Calculated by a support vector machine (Support Vector Machine) algorithm, for example: multiple images with gravure areas can be collected in advance, such as: training a support vector machine, you can get a classifier to calculate which belongs to it A type of ratio.

It should be understood that the probability of the printing area (such as the probability of printing) can also be based on a number of known gravure areas and a number of known non-gravure areas in a plurality of images with a K-nearest neighbor (K-nearest neighborhood algorithm and a Gaussian model (Gaussian model) calculation, such as: these printed areas and non-printed areas are scattered in space, when a sample enters, you can calculate the surrounding and its nearest k sample points, when the most recent sample The points belong to one of the more categories, then belong to that category, and then use the Gaussian distribution to calculate the probability value of this category.

It should be understood that the printing area probability can also be calculated by a convolutional neural network algorithm based on several known gravure areas and several known non-gravure areas in several images The results, such as: using a large amount of data and training a deep learning network for classification, can obtain a network with the ability to distinguish whether it is printing area data.

As shown in Figure 1, the printing step S6 can be performed by the control device according to each target The gravure coefficient of the target block extracts at least one target block as at least one gravure block, for example: extracts the target block whose gravure coefficient is greater than a threshold (such as a value between 0.5 and 0.9) as the gravure Block.

In one embodiment, as shown in FIGS. 1 and 3, in the fetching step S6, the control device may extract at least one target block B1 as the target block B1 according to the gravure coefficient of each target block B1 and the two-dimensional coordinates At least one gravure block, and at least one mark is generated in the image, such as various borders, various crosses, or other symbols, so as to facilitate user identification.

The following is an example to illustrate the experimental process of the above embodiment of the present invention, taking the gravure text on the steel embryo as an example, but not limited thereto. During the execution of the scanning step, imaging step, partitioning step, analysis step, conversion step, and printing step, the value of the difference conversion function G ( Sim ( x,y )) of the local block of an example image can be obtained, such as Table 1 shows.

For example, as shown in Fig. 4a, corresponding gravure blocks are shown in one image (the area of multiple blocks framed by thick lines in the figure, G ( Sim ( x,y ) ) High (that is, low correlation (that is, the similarity between the blocks is low)). Then, as shown in Figure 4b, a simple image processing method can also be used to remove the "low probability of occurrence of gravure" blocks, For example: a single block that is not connected to any block (as shown in the block marked with a solid cross "╳"). Then, as shown in Figure 4c, if the probability of using the above unprinted area V ( x,y ), the geometric shape can also be used to determine whether it is possible to be a gravure block (such as a block framed by a dotted line in the figure), if the block is too large or too small, it belongs to the object to be excluded. As shown in Figure 4d As shown, only the G ( Sim ( x,y )) = 0.5127, 0.3874 of the second block is marked as an example, and the G ( Sim ( x,y )) function value of the remaining blocks can refer to Table 1; if used In the above embodiment with printing area probability V ( x, y ), only blocks with low correlation (that is, low similarity between blocks and high difference between blocks) may be multiplied by the printing area probability V ( x,y ), there is no need to calculate the V ( x,y ) of all blocks in the whole picture, which can reduce the amount of data calculation. The probability of the printing area V ( x,y ) is too small (such as below a threshold) Blocks ( V ( x,y )=0.2106 as shown in Figure 4d can be considered excluded as gravure areas. Therefore, as shown in Figure 4e, it can be further excluded that the probability of some printing areas V ( x,y ) is too small And consider excluding the block that is the gravure area (the block marked with a dotted cross "╳" in the figure). Finally, as shown in Figure 4f, at least one gravure block can be obtained (the dotted line in the figure) Exclude the blocks with dashed cross symbols in the framed area).

In addition, since the gravure block usually has clustering, this feature can also be used to geometrically determine which gravure block in a plurality of blocks is the printing area, for example: a single block is more likely to be a printing area Low, can be excluded.

Through the above steps, the above embodiment of the present invention can automatically find the parts with high probability of gravure on the surface of the blank (such as steel embryo) to form the gravure area, so as to further obtain the characters of the gravure area by using text recognition technology Or the content of the pattern, the above embodiment of the present invention can also perform parallel operations, which has the effects of good accuracy and high processing efficiency. Compared with the conventional method, it can improve the efficiency of blank recognition and reduce the error of human eye recognition.

Although the present invention has been disclosed in preferred embodiments, it is not intended to limit the present invention. Anyone who is familiar with this skill can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the present invention The scope of protection shall be as defined in the scope of the attached patent application.

S1‧‧‧Scanning steps

S2‧‧‧Imaging steps

S3‧‧‧Partition steps

S4‧‧‧Analysis steps

S5‧‧‧Conversion steps

S6‧‧‧Printing steps

Claims (8)

A method for extracting gravure on the surface of an object. The method is executed by an electronic system. The electronic system includes a control device, a storage device, and a scanning device. The control device is electrically connected to the storage device and the scanning device. The device stores at least one instruction that enables the control device to perform actions and data and generated data required for the execution process. The method includes: a scanning step, the control device causes the scanning device to scan an object; an imaging Step, the control device or the scanning device generates an image based on a scan result; a partitioning step, the control device divides the image into two blocks in two dimensions; an analysis step, the control device The blocks are performed separately: the block is set as a target block, the target block is adjacent to several peripheral blocks, and the similarity between the target block and each peripheral block is calculated; a conversion step, Each target block is separately controlled by the control device: the similarity between all blocks of each target block is respectively introduced into a difference conversion function to generate the difference between several blocks of each target block, according to the The difference between the blocks generates a total value, which is calculated as an average value based on a number of the peripheral blocks, as a gravure coefficient for each target block; and a printing step, which is controlled by the control device Extract at least one target block as at least one gravure block according to the gravure coefficient of each target block. The method of extracting the surface gravure of the object as described in item 1 of the patent application scope, wherein in the conversion step, the control device is based on the difference between the blocks and the The product of the probability of a print area of the target block produces the total value. The method for extracting gravure on the surface of an object as described in item 2 of the patent application scope, wherein in the conversion step, the probability of the printing area is based on a known gravure area and a known non-gravure area in a plurality of images with a support Calculated by vector machine algorithm. The method for extracting gravure on the surface of an object as described in item 2 of the scope of the patent application, wherein in the conversion step, the probability of the printing area is based on the number of known gravure areas and the number of known non-gravure printing in several images The area is calculated by a K-nearest neighbor algorithm and a Gaussian model. The method for extracting gravure on the surface of an object as described in item 2 of the scope of the patent application, wherein in the conversion step, the probability of the printing area is based on the number of known gravure areas and the number of known non-gravure printing in several images The area is calculated by a convolutional deep learning network algorithm. The method for extracting the surface gravure of an object as described in item 1 of the scope of the patent application, wherein in the analysis step and the conversion step, the control device calculates the inter-block similarity of each target block according to the order of the target block and The degree of difference between blocks. The method for extracting the surface gravure of an object as described in item 1 of the patent application scope, wherein in the analysis step and the conversion step, the control device calculates the inter-block similarity and inter-block difference of each target block in parallel. The method of extracting gravure on the surface of an object as described in item 1 of the scope of the patent application, wherein in the partitioning step, the control device gives a two-dimensional coordinate to each block; and in the fetching step, the control device The gravure coefficients and the two-dimensional coordinates of each target block extract at least one target block as at least one gravure block, and generate at least one mark in the image.

Images