DE10040614A1 - Method for automatic detection of a directional structure - Google Patents

Abstract

In order to be able to automatically identify the location and orientation of a directed structure in a digital image, the invention provides a method in which the digital image is subdivided into preferably interconnected blocks. For each block, at least one texture characteristic is determined, which represents a measure for the structuring of the blocks along predetermined predominant directions, and parameters are determined from these texture characteristics. Said parameters indicate the location and the orientation of a directed structure in the digital image.

Description

The invention relates to a method for the automatic detection of a Location and orientation of a directional structure in a digital Image, especially for determining the location and orientation of a barcodes contained in the digital image.

There are different methods of detection and localization directed structures known in pictures. For example there is Procedures involving an image by using an operator (e.g. a Laplacian-of-Gaussian operator, or LoG operator for short) second image is calculated, the pixel values of which are not the brightness, but represent the local contrast at the location of the respective pixel. In in a second step, this second image is then examined, whether it's along substantially straight and essentially parallel Lines contain pixels that have high values.

A disadvantage of such methods is the high computational effort that complex computing units or long evaluation times required. Nice The use of an operator can be used for many applications be elaborate. This applies even more to the subsequent one Evaluation of the second image.  

Other methods determine that within local image areas two-dimensional autocorrelation function (AKF) or that two-dimensional Fourier spectrum. The values of the AKF or the Fourier spectra are then processed to make them Information about the structuring of the related areas receive. The computational effort for these methods is also very high.

The addressed directed structures in pictures can also Be barcodes. These barcodes are special optical codes that for example from parallel bars (bars) of different thickness consist of gaps (spaces) of different thicknesses are separated from each other. The sequence of the thicknesses of the bars and spaces represent the information stored in the barcode. Barcodes can be opened Goods packaging and labels can be printed, or for direct Marking of various products can be used. On The key feature is the machine readability of barcodes. This means that it is possible with special equipment (barcode reader) is to recover the information stored in a barcode.

There are currently basically two for reading barcodes different classes of procedures.

In one class, scanning the template with a Laser beam formed a time signal from a subsequent one Computing unit is evaluated. Here one speaks of one  so-called 1D process. For readers that use the 1D process the barcode can only be read if the Laser beam cuts the barcode across the bars and spaces. If the relative position of the barcode and reader is not a priori is fixed, but a reading of the code should nevertheless be made possible (regardless of location), it is necessary to pass the template along different Scanning directions several times (omnidirectional reading). According to the device, this is usually done with moving mirrors achieved that direct the laser beam so that the original at different Locations and along different directions. adversely is the need to use moving parts, which causes among other things, restrictions on the minimum size and the Tool life result.

In the other class, the scene is one with an electronic one Sensor equipped camera and recorded on the sensor digitized projected image. The digital image is then used for evaluation transmitted to a decoding unit (2D method). An advantage of 2D The process consists of moving the barcode reader to Parts can be omitted. Another advantage is that in the Decoding unit a digital image is present, which is an image of the Contains barcodes.

The decoding unit can optionally on the individual pixels of the Access picture, which makes the class of possible decoding methods  is significantly expanded. In particular, the entire repertoire of Methods of digital image processing in the sense of preprocessing of the image can be used. This can cause errors or Code malfunctions that occur during the printing process, for example can compensate within certain limits.

With the 2D methods for code decoding one can differentiate whether a determination of the location and orientation of the barcode in the image is carried out (localizing procedure) or not.

Localizing methods are particularly advantageous if the Location and / or orientation of the barcode within the field of vision the camera cannot be predicted with certainty and the barcode should also be read (omnidirectional reading).

With non-localizing methods, the ability to omnidirectional reading by evaluating the image along a Variety of virtual scan lines. So the picture becomes a 1D Signal formed by the gray levels of the image at different locations read out along different directions. The so educated 1D signal is thereupon fundamentally similar procedures evaluated like a time signal generated by a laser scanner. It is disadvantageous that the gray values of the image along a Large number of lines can be evaluated and thus on the one hand the  Computational effort multiplied and on the other hand the risk of Misreading is increased.

In a known method, which has these disadvantages by a local and Determination of the position of the barcode is avoided in the direct Additional markings are placed around the barcode. This Markings are made in a first step by one modified reader recognized. Since the markings are in one predetermined distance to the barcode is with the localization the markings also localized the barcode. In this procedure is disadvantageous that it requires printed markings. This one Markings are not part of a generally binding standard are not used consistently.

The additional markings also require an additional one printed area. Especially when marking goods (EAN / UPC) the printable area is valuable as an advertising medium, which is why the Unable to enforce the use of additional markings.

Since barcodes are made up of parallel bars, they pose in the sense structures oriented towards digital image processing and it is in principle possible, those originating from digital image processing Procedures for the localization of directed structures (texture analysis) too to locate barcodes. These procedures require however the use of special operators (e.g. LoG operator),  the determination of two-dimensional correlation functions or the Determination of the two-dimensional Fourier spectrum, i.e. the Application of computationally complex algorithms and are therefore for Localization of barcodes cannot be used economically.

It is therefore the object of the present invention, a method to provide, which is for the automatic detection of a location and an orientation of a directional structure in a digital image, in particular for recognizing one contained in a digital image Barcodes, as well as at least approximately determining the location and the orientation of this barcode can be used economically.

According to the invention, the above object is achieved in that in one Method for automatic recognition of a place and one Orientation of a directional structure in a digital image, in particular to determine the location and orientation of an in the digital image contained barcodes, the digital image in preferably contiguous blocks is divided and for each Block at least one texture feature is determined, which is a measure of the structuring of the blocks along given preferred directions represents, and parameters are determined from the texture features that the location and orientation of a directional structure in the digital Specify picture.  

The image is advantageously a camera-based one Code reader captured image of a real scene, the under contains a barcode. Under an electronically available or digital image becomes a matrix of image elements (Pixels) Compound object understood, with each pixel a location (usually in the form of a row index and column index) and a value characterizing the brightness at the location of the pixel (value of the pixel) assigned. But it can also be generated in other ways electronic form of available images with the presented method are analyzed automatically.

It is particularly advantageous in the method according to the invention that the computing effort compared to known localization methods directed structures, reduced many times over. The application This method of localizing barcodes is done by this reduced computing effort very interesting.

Another advantage of the method according to the invention is that no additional markings are required which with conventional methods for localization and Orientation determination for the barcode were required. Thus also no additional area required for the markings, which means this area is available as a possible advertising space.  

Another advantage of the method is that after the localization only a single line needs to be evaluated.

It also reduces the focus on those areas of the image that Most likely, barcodes contain the risk of Misreadings.

To locate a directional structure, the digital image is in Areas divided, the so-called blocks. It is advantageous if the blocks are contiguous and rectangular in shape, preferably have a square shape. For each of these Blocks, a predetermined number of first direction sensitive Features, the so-called "features of the first order" calculated. These "features of the first order" are also referred to as "first." Texture features ".

The calculation rule for these "first order characteristics" is described by so-called feature functions. A The feature function is an assignment rule that is assigned to a block uniquely assigns a number. All feature functions are designed so that the higher the values assigned to a block the larger the gray value differences of the block along one predetermined preferred direction. So it becomes a sentence of Feature functions are set, with each of the feature functions a preferred direction can be assigned to this sentence, along which  an image block is evaluated. The feature functions are like this designed that the associated preferred directions the entirety of all Cover directions as evenly as possible.

In particular, the feature function is chosen so that two arbitrarily selected different feature functions too have different preferred directions. Thereby a "characteristic first order "by applying a feature function on a Block determined. So you get for each block and for each Feature function a value.

It is advantageous if the ratio of the number of pixels per block to that Number of feature functions is significantly greater than 1, because then the Number of features per image is significantly less than the number of Pixel of the image, which increases the complexity of the following Calculation significantly reduced.

In order to save additional computing time and memory access time to win, it is advantageous to use only a part of the pixels one Evaluate blocks.

The feature function can also be seen as one Imagine "measuring device", the "activity" of an image or Measured image section along the preferred direction. The texture feature then corresponds to the "deflection" of the direction-sensitive measuring device "Characteristic function". If a block contains a directed structure, see above  becomes a texture feature that is essentially parallel to the structure runs, have a relatively low value. A feature that in runs essentially perpendicular to the structure have a relatively high value. A directed structure within a Blockes are therefore detected in two different ways.

On the one hand, a directional structure expresses itself through high values of one Feature function whose preferred direction is substantially perpendicular to Structure stands. On the other hand, a directional structure manifests itself low values of a feature function whose preferred direction in runs essentially parallel to the structure.

This results in two options, a directional structure recognizable in a block. On the one hand you can get particularly low ones Values of a texture feature as an indication of the existence of a directed structure along that belonging to the texture feature Use preferred direction, on the other hand you can use particularly high ones Values of a texture feature as an indication of the existence of a Use the directed structure perpendicular to the preferred direction.

In order to eliminate the influence of undirected structures, the methods according to the invention each consisting of two “features of the first order”, whose feature functions differ from one another essentially distinguish vertical preferred directions, a new "second Texture feature "can be determined. The connection of two  "First order features" is then designed so that the resulting "second texture feature" takes high values when the value of one "First order characteristic" high and at the same time the value of the other "First order feature" is low. So through Offsetting two "characteristics of the first order" certain "second" Texture feature "is also referred to as" second order feature "in the following. called.

A "second order characteristic" therefore takes on a high value if the block along the preferred direction has a high activity and at the same time, a low activity perpendicular to the preferred direction having. In this way it is achieved that neither structurally poor Areas still undirected structured areas at a high value of a "second order characteristic" and thus to a misinterpretation (supposed detection of a non-existent directed structure) being able to lead.

In the following, a "feature of the second order" is a feature understood, the value of which is a measure of the probability of The presence of a directional structure perpendicular to a preferred direction represents, regardless of whether it is by offsetting two "characteristics first order "or calculated in a single step.

The "second texture features" belonging to the same direction are entered in so-called feature cards, each on  the location of the associated block of the input image. The feature map is a reduced image, the pixels of which are not more the brightness of the original image, but the probability for the presence of a directional structure of a certain direction specify.

It is advantageous that not every pixel of the original image own characteristic value is assigned, but the different ones Characteristic values belong to the blocks and thus the characteristic card is correspondingly smaller than the original image.

It is also advantageous that the evaluation only relates to the previous one certain features, but not based on the gray values of the image. This allows a significantly lower computing effort to be achieved than for example, an image created with a LoG operator. This is because on the one hand due to the data reduction (the number of features is less than the number of pixels in the input image), but on the other hand this is also because the feature functions are constructed so that they Conclusions about the local orientation of the individual image areas allow.

These advantages make evaluation results more efficient and to achieve more economically. Stand as a result of this procedure Parameters are available that make a statement about the location and the Orientation or an angular position of a directional structure in one  specify digital image. Furthermore, a Probability for the existence of a directed structure determined become.

One embodiment provides that those "features second" Order "to zero in each feature card, whose Characteristic values are below a suitably chosen threshold, and only use those characteristics for further evaluation, whose value is really positive. The height of the threshold has essentially Influence on detection sensitivity.

With this method, it is advantageous to choose a low threshold, if you also detect directional structures with low contrast would like to. Conversely, it is advantageous to choose a high threshold if the structure to be recognized has a high contrast (e.g. with a perfectly printed barcode), but at the same time with high contrast textured disturbances in the environment of the sought directed structure.

First, a feature card is created from all feature cards selected, which with a high probability the existence of a directional structure in the image. It is also beneficial to those Select feature cards whose values have the largest mean exhibit. The "best" feature card selected in this way continues examined to confirm the existence of a directed structure or  to discard. The location of an existing structure in the picture can be determined by determining a position parameter (e.g. mean value, weighted mean, median) of the feature map can be estimated. The angular position of the directional structure is roughly derived from the preferred direction assigned to the selected feature card. A The more precise estimate of the orientation of the structure is given by the Use of a regression method from the feature map determined.

It is advantageous for the determination of the first order feature card a block-wise evaluation of gray value differences of Pixel pairs along the chosen preferred direction, because this Method versus the formation of the two-dimensional Autocorrelation function or that of the two-dimensional Fourier spectrum with significantly less computing effort is implementable.

In order to further minimize the computing and time expenditure, it is advantageous when evaluating the texture features first of all Select preferred directions, the associated texture features of which best explain with the existence of the directed structure in the picture to let. With this method, the data overhead can still be to reduce.  

Further advantages, goals and characteristics of the present invention will become apparent illustrated with the following explanations attached drawing, in which the detection method is shown as an example.

It shows

the figure is a schematic representation of the detection method based on an embodiment.

In the embodiment shown in the figure, one is Image acquisition unit connected to a computing unit. The Image acquisition unit delivers a digitized image of size 640 × 480 pixels with typically 256 gray levels to the computing unit.

The image is then first divided into 20 × 15 non-overlapping blocks Size 32 × 32 pixels divided. Be further for each of these Blocks four "features of the first order" determined. The Calculation rule for each "first-order characteristic" by described a feature function.

Each feature function uniquely assigns a number to each image block and is defined by two parameters, namely by

• 1. a so-called clique vector (Delta_x, Delta_y)
• 2. an increment DSAMPLE.

The components of the clique vector are integers (positive or negative), and the increment DSAMPLE is a natural number.

To determine a first-order feature, all pixel pairs ((x ₁ , y ₁ ), (x ₂ , y ₃ )) are formed which simultaneously meet the following conditions:

1. they lie completely within the block under consideration,

2. x ₁ -x ₂ = Delta_x and y ₁ -y ₂ = Delta_y,

3. x ₁ and y ₁ are divisible by DSAMPLE.

Condition 2 states that the pixel pairs are different

distinguish that on the one hand the connecting line between the Pixels of the pair along those given by the clique vector Preferred direction runs and that on the other hand the distance of two Points of a pair of points is constant.

Condition 3 allows only part of the condition 2 within of a block to select possible pairs of points and thus enables a reduction in the computing effort to determine the respective Texture feature.

The "first order characteristic" is determined as the next step. there for each pair of pixels that meets the above conditions, the Absolute amount of the difference or the square of the difference between the  Gray values of the two points determined. Then they are like this values obtained summed.

In the implementation example, four clique vectors are used, namely
c ₁ = (3.0) (this is the horizontal preferred direction),
c ₂ = (0.3) (this is the vertical preferred direction),
c ₃ = (2.2) (this is the diagonal preferred direction),
c ₄ = (-2.2) (this is the antidiagonal preferred direction).

DSAMPLE is set to 4, which results in savings Computing time of approximately 16 compared to a DSAMPLE value of 1 results.

In a next step the "first order characteristics" in so-called feature cards entered.

In the implementation example, the "features thus obtained" first order "in four (number of clique vectors) feature cards entered, each size 20 × 15 (number of blocks = number Features per clique vector).

In the following, the "feature cards of the first order" are designated T_n (x, y), n taking the values 1 to 4, x the values 1 to 20 and y the values 1 to 15. For example, T_2 (3, 4) denotes the "feature of the first order" for the third block from the left in the fourth line for the clique vector c ₂ . The feature T_2 (3, 4) in this example indicates the activity of the block at position (3, 4) along a vertical direction.

In a further step, the "feature cards become first Order "determines the" feature cards of the second order ".

To determine a "second order feature card" M_i, a "first order feature card" T_i is offset against another "first order feature card" T_j. The calculation is carried out with a feature map T_j, the clique vector c _{j of which is} perpendicular to the clique vector c _{i of} the "feature map of the first order".

In the example, c ₁ and c ₂ or c ₃ and c _{4 are} perpendicular to each other. The type of calculation is determined by two parameters FACTOR and THRESHOLD and is carried out in blocks as described below. First, the "second order feature cards" are determined as follows:
M_1 (x, y) = T_1 (x, y) - FAKTOR.T_2 (x, y),
M_2 (x, y) = T_2 (x, y) - FAKTOR.T_1 (x, y),
M_3 (x, y) = T_3 (x, y) - FAKTOR.T_4 (x, y),
M_4 (x, y) = T_4 (x, y) - FACTOR.T_3 (x, y).

The next step is to determine the threshold that is set by the depends on the respective application. The threshold is in essentially from the selected block size and the gray value resolution of the original image. In this step all entries in all feature cards set to zero if they are smaller than this one Are threshold.

Typical values for the factor are in the range of 1.5. , , 2.5. The feature cards are then sorted. This happens in such a way that for each feature card the number of non-zero entries is counted and the feature cards according to this number in descending order Order. Thus the first of the feature cards in the list most non-zero entries and is therefore considered first candidate considered.

The center of gravity of the feature cards to be considered (the first moment in the x and y directions) is then determined. The center of gravity (x _s , y _s ) has real-valued coordinates in the range x _s = 1. , , 20, y _s = 1. , , 15 and is rounded to integer values round (x _s , y _s ).

If the feature card under consideration contains a non-zero value at round (x _s , y _s ), the feature card is continued as the current map.

If, on the other hand, the feature card under consideration contains a zero instead of round (x _s , y _s ), the center of gravity of the next feature card is determined from the list. It is then checked whether this card contains a non-zero value at the point of the center of gravity.

In this way, all candidates on the list are processed until either a map was found that was at the point of focus contains a non-zero value or until the list is unsuccessful was processed.

In the latter case, the procedure is ended because there is no directed structure was found.

Otherwise, the center of the barcode (based on the size of the original image) is determined from the center of gravity (x _s , y _s ) of the current feature map by scaling with the block size. A regression line running through the center of gravity (x _s , y _s ) is determined for the feature map by linear regression.

This straight line runs essentially vertically through the directed structure. If the directional structure is now a barcode that This line can be decoded as a virtual scan line be used as they are very likely to contain the barcode completely cuts.

The procedure described here for the automatic detection of a Location and orientation of a directional structure in a digital Image is characterized by the data reduction in the original image  contained data, so that a significantly lower computing effort must be operated in order to localize a directed structure can. It is advantageous that the digital image in adjacent blocks is divided and these blocks through Feature functions are described so that not every single one Pixels in the input image are included in the calculation, but only the number of the texture features.

It is also advantageous that the possibility of incorrect readings thereby is reduced that the procedure allows to focus on those Focus areas of the image that are highly Barcode included.

Claims (10)

1. A method for the automatic detection of a location and an orientation of a directional structure in a digital image, in particular for determining the location and the orientation of a bar code contained in the digital image, characterized in that the digital image is subdivided into preferably coherent blocks, and at least one texture feature is determined for each block, which represents a measure for the structuring of the blocks along predetermined preferred directions, and parameters are determined from the texture features that indicate the location and the orientation of a directional structure in the digital image. 2. Detection method according to claim 1, characterized in that that the blocks have a rectangular, preferably square shape exhibit. 3. Detection method according to one of the preceding claims, characterized in that several "first Texture features "are determined, the higher the value assume the larger the sum of the absolute amounts of the Gray value difference of pixel pairs fails, whereby the Pixel pairs for each "first texture feature" from pixels are formed which lie within the block under consideration,  their connecting line along the "considered first" Texture feature "assigned preferred direction, and their Distance from a specified value or a value from a set set of values. 4. Detection method according to claim 3, characterized in that that only a portion of the pixels within the viewed Blocks whose connecting line along the assigned Preferred direction and their pixel spacing one specified value or a value from a specified quantity of values correspond to be evaluated. 5. Detection method according to one of the preceding claims, characterized in that to determine the value of a "second texture feature" of a block the values of two "first texture features" of the same block that are a measure of the structuring of the block along two essentially perpendicular to each other Specify preferred directions. 6. Detection method according to one of the preceding claims, characterized in that the calculated "first Texture features "are evaluated. 7. recognition method according to one of the preceding claims, characterized in that when evaluating the  Texture features first selected that preferred direction whose associated texture features best match the Existence of a directed structure in the picture reveals and at then only those texture features that correspond to the selected preferred direction belong to the determination of the Parameters that determine the likelihood of existence directional structure and / or the location of a directional structure and / or the position of a directional structure in the digital image specify to be used. 8. recognition method according to one of the preceding claims, characterized in that the image is in overlapping areas is divided. 9. Detection method according to one of the preceding claims, characterized in that to adjust the Detection sensitivity a threshold when evaluating the texture features is determined. 10. recognition method according to one of the preceding claims, characterized in that the parameters are a Probability of the existence of a directed structure in specify the digital image.