Classifications

• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T7/00—Image analysis
  • G06T7/0002—Inspection of images, e.g. flaw detection
  • G06T7/0004—Industrial image inspection
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F18/00—Pattern recognition
  • G06F18/20—Analysing
  • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
  • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
  • G06F18/2132—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on discrimination criteria, e.g. discriminant analysis
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F18/00—Pattern recognition
  • G06F18/20—Analysing
  • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
  • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
  • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F18/00—Pattern recognition
  • G06F18/20—Analysing
  • G06F18/24—Classification techniques
  • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
  • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V10/00—Arrangements for image or video recognition or understanding
  • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
  • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V10/00—Arrangements for image or video recognition or understanding
  • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
  • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
  • G06V10/7715—Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/20—Special algorithmic details
  • G06T2207/20021—Dividing image into blocks, subimages or windows
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T7/00—Image analysis
  • G06T7/0002—Inspection of images, e.g. flaw detection
  • G06T7/0004—Industrial image inspection
  • G06T7/001—Industrial image inspection using an image reference approach
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
  • G06V2201/06—Recognition of objects for industrial automation

Definitions

• the invention relates to the field of tire manufacturing and more particularly the control of the internal appearance of the surface of a tire using automatic means to assist the operators responsible for these operations.
• the gray level image of the inner surface of the tire is obtained using conventional means. These means may be of the type making it possible to obtain a gray level image.
• the present invention is an improvement of the above method whose purpose is to improve the calculation time required for the image processing by taking advantage of the specificities of the internal surface of a tire and the relevance of the results obtained.
• the baking membrane comprises in fact raised patterns such as regular striations forming a given angle with the radial direction, or still mottles forming non-periodic structured patterns of random type. These patterns are reproduced hollow on the inner surface of the tire.
• the object of the method according to the invention is to provide solutions to all the problems posed above and comprises the steps in which:
• the gray level start image of said surface is captured and transformed into an orthonormal space in which the abscissa axis (OX) represents the circumferential direction, and the ordinate axis (OY) represents the radial direction ,
• the starting image is transformed using a predetermined series of filters to obtain a multivariate starting image.
• the distribution of the imagettes containing an anomaly with respect to the distribution of the imagettes containing no anomalies is evaluated in order to assess the separating power of the descriptor under consideration.
• the series of filters is formed from morphological operators such as dilation, erosion, opening, closure, or residues of one or more of these operations.
• the tessellation of the images is made in the circumferential and radial directions so that said thumbnails have the shape of a square.
• the sides of the square of the image are preferably formed by a number of pixels corresponding to a length of between 0.01 * ⁇ and 0.1 * ⁇ , ⁇ being the seat diameter of the tire to be controlled.
• the data analysis tool is preferably chosen from one of the following factor analysis tools:
• PCA principal component factor analysis
• the mean per line or column, the mean, the variance, the standard deviation, the extremal values, the deviation at the maximum or minimum amplitude, the average positive or negative crossing of the gray levels in the circumferential or radial direction.
• the reduced factorial spaces of each of the descriptors whose separating power has been evaluated as relevant are grouped into a reduced common factor space.
• the method according to the invention comprises, as seen above, a learning phase during which we will select descriptors relevant to the anomalies that we seek to detect, and a control phase during which, equipped with these relevant descriptors will analyze the inner surface of a tire to determine the presence or absence of anomalies.
• the predetermined series of filters is applied to the starting image of the inner surface of the tire to be controlled to obtain a multivariate starting image
• the multivariate starting image is cut according to the predefined tiling in the axial and circumferential directions, so as to obtain multivariate images of the inner surface of the tire to be inspected,
• each of the multivariate images is transformed into one-dimensional vectors, using each of the selected descriptors, so as to obtain a simplified multivariate image of the inner surface of the tire to be controlled,
• the classifier is constructed using one of the following analysis tools:
• FIGS. 1 to 6 The following description is based on FIGS. 1 to 6 in which:
• FIG. 1 represents a gray level image of the inner surface of a tire containing an anomaly
• FIG. 2 represents a schematic view of the image of the inner surface unwound in an orthonormal frame
• FIG. 3 represents a schematic view of the multivariate image obtained after application of a series of filters
• FIG. 4 represents a schematic view of the division of the image of FIG. 3 into multivariate images
• FIG. 5 represents a schematic view of the multivariate image obtained after transformation of each of the multivariate images by a descriptor
• FIG. 6 represents a 3D graph of the position of the images in a reduced factor space.
• Figure 1 shows a gray level image of the inner surface of a tire. There is clearly an anomaly 1 related to an irregularity of the son of the carcass reinforcement ply which is detached from the structured bottom comprising streaks 2 or mottles 3.
• the gray level can simply be the result of the level of reflection of the light intensity as is the case for Figure 1. In which case we will use a traditional camera.
• polynomial interpolation can be performed on the gray level averages by treating separately the row profiles and the column profiles.
• the compensation of the drift is obtained by subtracting row by row and column by column the values of the polynomial obtained.
• This method smoothes the image while retaining strong local features.
• the application of the method according to the invention does not differ according to whether one is dealing with a two-dimensional gray level image, or with an image whose gray levels reflect the altitude points of the surface.
• the image resulting from the sensor is transformed into a picture unwound in an orthonormal coordinate system OXY so that the circumferential coordinates are carried by the abscissa axis, and the coordinates radial are carried by the ordinate axis OY.
• the gray-level start image is denoted by /.
• tires are selected comprising zones where there are one or more anomalies previously identified and identified. It will also be noted that it is possible at this stage to manually embed zones comprising anomalies taken from images made on different tires.
• the next step of the method consists in transforming the initial image using a predetermined series of N filters F, F 2 f , i 7 , so as to extract the information contained in the texture of the image.
• filters can be indistinctly morphological operators.
• morphological operator is meant here operations such as erosions or expansions which consist, for each point of an image, in finding the minimum value or the maximum value of gray level of the neighboring points included within a structuring element, of given shape and surface, defining a neighborhood of the point to be analyzed. For an erosion the value at this point then becomes the minimum value, and for a dilation the value at this point becomes the maximum value. The combination of erosion followed by expansion is called opening, and the combination of dilation followed by erosion is called closure.
• These operations can be combined in series. Similarly, one can operate on the residues of these operations to perform operations.
• small linear structuring elements oriented in the circumferential direction OX or in the radial direction OY will be chosen.
• V f (x t ) is the response of one pixel to the collection of filters:
• the next step is then to reduce the size of this multivariate image.
• a tiling is defined in the axial and circumferential dimensions delimiting the boundaries of small sized images in the 2D space formed by the image
• the size of the meshes of the paving is adjusted in a manner adapted to the size of the tire to be inspected, and it must be defined so as to contain enough information not to be short-sighted with respect to the anomalies that the we try to detect.
• a Representative Elemental Surface (SER) is evaluated on the healthy background which corresponds to the average surface of the finest textured elements such as the mottles 3 illustrated in FIG. This result can be obtained by calculating, using a covariance function, the characteristic size of the elements of the texture, which is expressed in the form of the mathematical expectation of the product of the image with its translated by a vector. h at which we subtract the average of the image squared.
• the thumbnails have a square shape of 256x256 pixels for the tire dimension mentioned above, which corresponds to a value between 0.01 ⁇ and 0.1 ⁇ , ⁇ being the seat diameter of the seat. pneumatic to control.
• the multivariate image is then in the form of the juxtaposition of I ' m x J' m images forming each channel F of the multivariate image.
• the multivariate thumbnail is n
• the following step of the implementation of the method according to the invention provides for associating with each of these multivariate images, a vector 1 D corresponding to the transformation of this multivariate image using a descriptor .
• the objective of this step is to reduce the size of the multivariate image in order to find the descriptors most likely to react to the presence of a surface anomaly without losing the information contained in the image to be analyzed. This operation concentrates the information contained in a thumbnail image for subsequent processing.
• a new simplified multivariate image is thus obtained, which is formed of the juxtaposition of the transformation by said descriptor of each of the thumbnails contained in each of the images F, F 2 f , T / obtained by application of the filters of the series of selected filters.
• D is the support space of a 2D image Ff of dimension I m x J m
• T a R, and T NL N (TxTx ... x T), where Z H is the dimension of the image space T 1 TM of the transformation of a thumbnail belonging to an image F. f , and N le number of filters x k e D, a pixel of the multivariate image obtained by transforming the imagettes into vectors.
• H w "(x k k) is therefore a descriptor of the multivariate image.
• the object of the invention is therefore to propose descriptors which have been judged to be the most efficient for analyzing the inner surface of tires.
• the descriptor will be chosen from among the following operators, the definitions of which will be briefly described below:
• the grayscale histogram of an image is a function that gives the occurrence (or frequency) of occurrence of each gray level in the image.
• H ⁇ ( 3 ⁇ 4 ) [H ⁇ (3 ⁇ 4), H ⁇ (3 ⁇ 4),
• the value of h varies from 0 to 60 pixels.
• the multivariate thumbnail associated with this descriptor is noted:
• G " P (3 ⁇ 4) G " * (3 ⁇ 4), G " * (3 ⁇ 4),, G " 3 ⁇ 4 (3 ⁇ 4) Radial covariance
• the value of h varies from 0 to 60 pixels.
• the multivariate image with this descriptor is noted:
• ⁇ ⁇ ( 3 ⁇ 4 ) ⁇ ⁇ ( 3 ⁇ 4) , ⁇ ⁇ (3 ⁇ 4),, ⁇ 3 ⁇ 4 (3 ⁇ 4) Averages and Variances by row or column in radial or circumferential directions
• PCA Principal Component Analysis
• n be the number of thumbnails and p the dimension of the variables (total dimension of the descriptor considered).
• Each data is represented in a space with p dimensions, the set of data forms a "cloud of n points" in R p .
• the principle of PCA is to obtain an approximate representation of the cloud in a subspace of smaller dimension k, by projection on well-chosen axes.
• the k principal axes are those that maximize the inertia of the projected cloud, that is, the weighted average of the squared distance squares. projected at their center of gravity.
• the axes defined by the ACP are orthogonal.
• the principal components are the n vectors whose coordinates are those of the orthogonal projections of the n elements of the cloud on the k principal axes. Unlike the initial variables, these new coordinates are two-by-two uncorrelated. It is possible to not consider that a reduced number of axes that represent a large part of the variance of the data (for example 90% of the total inertia). The choice of the number of axes to consider can be done in different ways
• FFA Correspondent factor analysis
• Figure 6 allows, as an example to view the result obtained.
• the chosen descriptor is the measure of the radial covariance of each of the thumbnails of the database.
• the Principal Components Analysis applied to these data allows to rank the importance of the factorial axes according to the inertia of the point cloud with respect to these axes.
• the inertias of the first factorial axes are as follows: 42.3%, 19.1%, 7.5%, 5.3%, 4.8%, 4.1%, 3.3%, 3.1%, 2.3%, 1, 9%.
• the analysis of the first 10 factorial axes can be considered sufficient because it contains 93.7% of the information.
• Figure 6 shows the projection of the images in the space of the first three axes of the PCA of the vertical covariance. Although these factorial axes only present 68.9% of the total inertia, we observe an interesting distribution of the images. Thumbnails without anomalies, represented by crosses, form indeed a fairly homogeneous group in this space, while thumbnails with defects, of the type illustrated in Figure 1 and represented by circles, have a much more sparse distribution. . In addition, there seems to be a hyperplane ⁇ separator between the two groups of images. This descriptor therefore seems to provide interesting information with regard to the classification of the images in two groups.
• the learning phase continues iteratively by applying the method described above, as many times as necessary, to a collection as wide as possible of images containing anomalies, so as to select the descriptors presenting the better separator power.
• the descriptors H, G, N selected it is possible to group the reduced factorial spaces obtained during the selection of a descriptor, in a single reduced common factor space.
• the grouping of the factor spaces into one is all the easier as in these spaces, the variables are normalized.
• the canonical metric of factorial spaces is the Euclidean metric. This metric is used in most classifiers.
• x - »c (x) [073] It is also possible at this stage to go directly to the classifier construction step, which consists of determining the areas of the factor space in which the pixels considered to form the image of the image are located statistically significantly. the anomaly. [074] The method then provides to construct a classifier that will detect the presence (or absence), as well as the position of an anomaly. The classifier makes it possible to isolate certain areas of the common reduced spectral space in which there are significantly cloud points corresponding to the spectral image of the imagettes comprising an anomaly, and the point clouds corresponding to the images not including abnormalities.
• a first method consists in applying an analysis method based on a linear discriminant analysis (LDA).
• LDA linear discriminant analysis
• This method of analysis aims to separate classes of points by hypersurfaces whose dimension is equal to the number of classes minus one, assuming that the distribution of points in a class is Gaussian. This works well in many cases, even if the points of the classes do not quite have a Gaussian distribution.
• the LDA can of course be used in multidimensional spaces.
• the calculation times can vary significantly depending on the choice of the kernel, but the SVM can model various problems.
• the gray level digital image is produced of the surface of the tire that is to be sorted, and this initial image is transformed into an orthonormal space in which the abscissa axis (OX ) represents the circumferential direction, and the ordinate axis (OY),
• the initial image of the inner surface of the tire to be controlled is applied to the predetermined series of filters F, F 2 f , F [to obtain a multivariate starting image.
• the multivariate starting image is cut according to the predefined tiling in the axial and circumferential directions, so as to obtain multivariate images of the inner surface of the tire to be controlled, and each of the channels of the multivariate images is converted into one-dimensional vectors. , using each of the descriptors H, G, N selected, so as to obtain a simplified multivariate image of the inner surface of the tire to be controlled.

Landscapes

• Engineering & Computer Science (AREA)
• Theoretical Computer Science (AREA)
• Computer Vision & Pattern Recognition (AREA)
• General Physics & Mathematics (AREA)
• Physics & Mathematics (AREA)
• Data Mining & Analysis (AREA)
• Artificial Intelligence (AREA)
• Evolutionary Computation (AREA)
• Databases & Information Systems (AREA)
• Life Sciences & Earth Sciences (AREA)
• Medical Informatics (AREA)
• General Health & Medical Sciences (AREA)
• Computing Systems (AREA)
• Multimedia (AREA)
• Health & Medical Sciences (AREA)
• Software Systems (AREA)
• Bioinformatics & Cheminformatics (AREA)
• Bioinformatics & Computational Biology (AREA)
• Evolutionary Biology (AREA)
• General Engineering & Computer Science (AREA)
• Quality & Reliability (AREA)
• Image Analysis (AREA)
• Image Processing (AREA)

Applications Claiming Priority (2)

Publications (1)

Family

ID=46924477

Family Applications (1)

Country Status (5)

Families Citing this family (8)

Family Cites Families (5)

• 2011
  • 2011-09-30 FR FR1158800A patent/FR2980735B1/fr active Active
• 2012
  • 2012-09-27 WO PCT/EP2012/069152 patent/WO2013045593A1/fr not_active Ceased
  • 2012-09-27 US US14/348,156 patent/US9189841B2/en active Active
  • 2012-09-27 EP EP12762640.6A patent/EP2761587A1/de not_active Withdrawn
  • 2012-09-27 CN CN201280047802.7A patent/CN103843034A/zh active Pending

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events