DE102007031215A1 - Welding seam inspection method for use during building of car body, in automobile industry, involves using thermal image to detect and evaluate welding seam with respect to different types of defects - Google Patents

Abstract

The method involves establishing a feature vector that represents a time profile of heat flows (11-13). The feature vector is used to ascertain two characteristic thermal images, which corresponds to minimum heat flow and maximum heat flow, respectively, through an object under test, from a series of thermal images. One of the thermal images is used for defect types, to detect and evaluate the welding seam with respect to different types of defects, where the characteristic thermal images serve as references for determining a suitable thermal image for the defect types.

Description

The The invention relates to a method according to the preamble of claim 1.

Especially The invention relates to nondestructive, imaging Testing methods of an object, eg. B. Thermography, X-ray, ultrasound or HDR examination. In these methods, the appropriate Radiation in the form of an image with an infrared, X-ray or high-dynamic-range camera or an ultrasound device, transferred to a processing unit and there manually or evaluated by image processing.

The thermography alone can provide the result images of various types, for example, a thermal image or an amplitude or phase image is obtained ( Theory and Practice of Infrared Technology for Nondestructive Testing / Xavier PV Maldague. - Jonh Wiley & Sons, Inc., 2001 ). Thus, the images obtained with different techniques have very different dynamics depending on the type of generation and the type of result. These lie, for example, in the value range from 4096 to 65536 gray scale levels, which corresponds to an intensity resolution of 12 to 16 bits, and can have both positive and negative values. Thus, the images described above differ significantly from an 8-bit dynamic image, which is well known for a visual control of its content has established. This plays z. For example, the key role in setting up an image processing system is to be parameterized step by step, with all of these steps being visually controlled by an operator.

Around Review images with different dynamic ranges and visually to be able to compare with each other, one needs a procedure with which a picture in the desired dynamics adequately can be converted to its content. In an industrial Application of this procedure is to be noted that an automatic Conversion of image dynamics ensured with a repeatable high quality becomes. Normally a won highly dynamic (eg 12-16 Bit) picture in a usual image processing 8 bit image converted.

As prior art, a method for dynamically converting an image based on two thresholds has been established. These absolute thresholds variously delineate an intensity range from the entire spectrum of a highly dynamic image ( High Dynamic Range (HDR) Vision / Berndt Höfflinger. - Springer, 2006, pp. 110-113 ). The detected intensity values are then converted to a lower dynamic range of typically 8 bits. Both a linear (stretch mode) and a non-linear (eg Rec. 709 mode) dynamic conversion can take place. The simplest and best known method of performing such a dynamic transformation is to use the minimum and maximum intensity values of the image as the lower and upper thresholds. This will use the entire intensity range of the captured image for the dynamic transformation of this image. The required absolute thresholds can also be determined empirically and used for all further recordings. However, this method causes individual outlier pixels to cause a false result.

Further Another method is known which involves a dynamic transformation of a Image on the basis of its histogram. at This method, known as CatEye, becomes the transformation function for the intensity conversion of an image alone from the frequency distribution of the intensity values of the original image. However, this means a rigid one Conversion method, which in a constantly varying Image scene fails.

In a further modification of this method, known as CatEye2, an intensity range is formed which delimits from the lowest or the highest intensity value that the captured image has, with the lower limit value I _{bot_thd} or the upper limit value F _{top_thd} . These limit values can be determined dynamically from the calculated histogram, wherein they cut off a predefined characteristic area _fraction F _{bot_thd} or F _{top_thd of} the entire histogram area from below or from above. In this method, however, the characteristic area _fractions F _{bot_thd} and F _{top_thd} undifferentiatedly _relate to the entire content of the recorded image, including different perturbations, so-called "dead pixels" and various artifacts, which results in the intensity areas _{corresponding to} at least one subject being examined Therefore, the dynamic transformation of the captured image will also be a false result, and the CatEye2 method will produce results that depend on the size of the object being examined and its background.

The dynamics of an image on which at least one object to be examined is located on a background should be converted so that its content is displayed as adequately as possible on the new image. This means that the information-relevant intensity ranges are determined from the content of the recorded image, which together form an information-relevant basis of the image content It is used to represent the recorded objects on their background, including their characteristic features, such as their edge, structure, as well as form and size, undistorted on the converted image.

outgoing This is the object of the invention, a method to create an adaptive dynamic transformation of an image in which the Content of a picture taken in a dynamic adequate presented in a different dynamic. Furthermore the procedure should be adapted to the different recording and evaluation methods behavior. It should also be flexible to let. This will make this process as universal and reliable identify.

The Solution to the problem arises from the characteristics of Claims 1 to 7.

According to claim 1, the method for the adaptive dynamic transformation of an image provides for the calculation of its histogram. From the histogram, the intensity range corresponding to at least one object to be examined and the intensity range corresponding to the background are determined. From each of these intensity ranges, the corresponding total area that lies between the histogram curve and the intensity axis is calculated. Then the absolute limits I _{obj_bot_thd} and I _{obj_top_thd} for the intensity range corresponding to at least one object to be examined and the absolute limits I _{bgrd_bot_thd} and I _{bgrd_top_thd} for the intensity range corresponding to the background are determined. These limit values delimit their information-relevant portion in each intensity range, whereby they separate a local intensity range from below or from above for each intensity range. The truncated local intensity _{regions having} the characteristic area _fraction F _{bot_thd} and F _{top_thd} , respectively, have individual outlier pixels, extreme intensity values, and random noise values, respectively. Thus, the information-relevant basis of the content of the recorded image is determined and the outlier pixels lying outside it will not exert any negative influence on the dynamic transformation of this image.

Since the calculated histogram curve can be considered as a combination of the corresponding Gauss normal distribution densities of the objects to be examined and their background as well as the interferences that have occurred ( Digital Image Processing / Bernd Jähne. - 4th edition - Berlin, Heidelberg, etc .: Springer, 1997 ), the characteristic surface _fractions F _{bot_thd} and F _{top_thd can be compared} with a value of the probability _integral of the normal distribution according to Gauss ( Mathematical Handbook / Granio A. Korn, Theresa M. Korn. - McGraw-Hill Book Company. New York, 1968 ). In general, they can each represent a 2.5% share. Thus, according to the invention, the remaining surface area of each intensity range will have an industrially acceptable value of 0.95. In a specific system, however, according to claim 2, the suitable sizes of the characteristic surface portions F _{bot_thd} and F _{top_thd} whose values are in the range [0-0.5] can be determined according to experience. Thus, a secure acquisition of the information-relevant basis of the content of the recorded image is guaranteed.

According to claim 3, the sizes of the characteristic surface portions F _{bot_thd} and F _{top_thd are set} independently of each other. This will allow a flexible and secure determination of the information-relevant base of the captured image.

For the same reason, according to claim 4, the magnitudes of the characteristic area _components F _{bot_thd} and F _{top_thd} for the intensity area corresponding to at least one object to be examined are F _{obj_bot_thd} and F _{obj_top_thd} and for the intensity area corresponding to the background, F _{bgrd_bot_thd} and F _{bgrd_top_thd is set independently} for each of these intensity _ranges .

According to claim 5, the intensity range, which is at least one corresponding to the examining object, and the intensity range, which corresponds to the background, on the histogram of the recorded Image determined using morphological filter. This ensures that the corresponding intensity ranges of the histogram in its overall course and thus independent of the random ones Malfunctions detected correctly and evaluated correctly become.

According to claim 6 can be the histogram from which the information base of the Content of the recorded image is calculated, from one or several areas of this image to be examined are formed. This will be a targeted and accurate capture at least of an object to be examined from the whole converted image guaranteed.

According to claim 7, images which are recorded in different wavelength ranges in one or in different dynamics can be combined into an image of desired dynamics. In doing so, the dynamics of each image obtained are converted into the desired dynamics according to the present method. The generated images are then fused to a common image using known image processing techniques. The image thus converted will allow both proper visual control and further automatic processing ben.

According to the invention the method described for both a reduction as well as for an increase in image dynamics become.

The details of the invention and its further features, applications and advantages are in the following embodiment with reference to the 1 explained. All described or illustrated features, alone or in any combination form the subject matter of the invention, regardless of their summary in the claims or their dependency and regardless of their formulation or representation in the description or in the drawing.

1 shows the inventive determination of the absolute limits I _{bgrd_bot_thd} and I _{bgrd_top_thd} and I _{obj_bot_thd} and I _{obj_top_thd} the Intensitätsbetreiche the histogram of the recorded image that correspond to at least one object to be examined and the background. These limits serve as image-specific thresholds for the dynamic transformation of this image.

As an example, a highly dynamic image can be taken on which an object and its background as well as the disturbances present on the image each have a local frequency maximum ( 1 ). The high dynamics of this image should be adequately converted to its content in a lower dynamics.

Despite the boundary conditions, the histogram ( 1 ) of the recorded image using the known image processing methods determine the Intensitäts Betreiche that correspond to the object to be examined and its background. According to the invention, this analysis is carried out with the aid of morphological filters and thus a trouble-free evaluation of the histogram curve (FIG. 1 ) guaranteed. From each of these intensity ranges, the corresponding total area that lies between the histogram curve and the intensity axis is calculated. Then in the intensity range ( 2 ) corresponding to the object to be examined, the lower one ( 4 ) as well as the upper ( 5 ) local intensity range containing noise or random noise values. These local intensity _ranges are each given a characteristic area _fraction F _{obj_bot_thd} or F _{obj_top_thd of} the intensity range ( 2 ) exhibit. In the intensity range ( 3 ), which corresponds to the background, the lower () 8th ) as well as the upper ( 9 ) local intensity _{range is} calculated with corresponding characteristic area _fractions F _{bgrd_bot_thd} and F _{bgrd_top_thd} . Thus absolute limits I _{obj_bot_thd} ( 6 ) and I _{obj_top_thd} ( 7 ) and I _{bgrd_bot_thd} ( 10 ) and I _{bgrd_top_thd} ( 11 ) determines which of the above-mentioned local intensity ranges ( 4 ) and ( 5 ) such as ( 8th ) and ( 9 ). The intensity ranges lying between these limit values together form the information-relevant basis of the content of the recorded image. Thus, an adaptive dynamic transformation of this image is guaranteed regardless of the image acquisition technique and the size of the object to be examined or its background and the interference that occurred by the contents of the recorded image is adequately represented in a different dynamic.

1

histogram curve

2

Intensity range, which corresponds to the object to be examined

3

Intensity range, which corresponds to the background

4

Lower local intensity range in the intensity range ( 2 ), which contains noises or random noise values

5

Upper local intensity range in the intensity range ( 2 ), which contains noises or random noise values

6

Absolute lower limit of the information-relevant share ( 12 ) of the intensity range ( 2 )

7

Absolute upper limit of the information-relevant share ( 12 ) of the intensity range ( 2 )

8th

Lower local intensity range in the intensity range ( 3 ), which contains noises or random noise values

9

Upper local intensity range in the intensity range ( 3 ), which contains noises or random noise values

10

Absolute lower limit of the information-relevant share ( 13 ) of the intensity range ( 3 )

11

Absolute upper limit of the information-relevant share ( 13 ) of the intensity range ( 3 )

12

Information-relevant proportion of the intensity range ( 2 )

13

Information-relevant proportion of the intensity range ( 3 )

QUOTES INCLUDE IN THE DESCRIPTION

This list The documents listed by the applicant have been automated generated and is solely for better information recorded by the reader. The list is not part of the German Patent or utility model application. The DPMA takes over no liability for any errors or omissions.

Cited non-patent literature

• - Theory and Practice of Infrared Technology for Nondestructive Testing / Xavier PV Maldague. - Jonh Wiley & Sons, Inc., 2001 [0003]
• - High Dynamic Range (HDR) Vision / Berndt Höfflinger. Springer, 2006, pp. 110-113 [0005]
• - Digital Image Processing / Bernd Jähne. - 4th edition - Berlin, Heidelberg, etc .: Springer, 1997 [0012]
• - Mathematical Handbook / Granio A. Korn, Theresa M. Korn. - McGraw-Hill Book Company. New York, 1968 [0012]

Claims (7)

A method of automatically dynamically converting an image, wherein a histogram of a dynamic image is calculated and used to adequately represent its content in another dynamic, characterized in that the absolute thresholds I _{obj_bot_thd} and I _{obj_top_thd} are the information _relevant portion ( 12 ) of the intensity range ( 2 ), which corresponds to at least one object to be examined, as well as the absolute limits I _{bgrd_bot_thd} and I _{bgrd_top_thd} , which _contain the information- _relevant portion ( 13 ) of the intensity range ( 3 ), which corresponds to the background, are determined dynamically, so that the limit value I _{obj_bot_thd} or I _{bgrd_bot_thd has} a characteristic area _fraction F _{bot_thd of} the histogram area of the intensity area ( 2 ) or intensity range ( 3 ) from below and the limit value I _{obj_top_thd} or I _{bgrd_top_thd} a characteristic area _fraction F _{top_thd of} the histogram area of the intensity area ( 2 ) or intensity range ( 3 ) from above, and thus an information-relevant basis of the content of the recorded image is determined, which is used for the dynamic transformation of this image. A method according to claim 1, characterized in that the sizes of the characteristic surface portions F _{bot_thd} and F _{top_thd are determined} according to experience. A method according to claim 1 and 2, characterized in that the sizes of the characteristic surface portions F _{bot_thd} ^and F _{top_thd are set} independently. Method according to Claims 1 to 3, characterized in that the magnitudes of the characteristic surface portions F _{bot_thd} and F _{top_thd} for the intensity range ( 2 ), which corresponds to at least one object to be examined, as F _{obj_bot_thd} and F _{obj_top_thd} as well as for the intensity range ( 3 ), which corresponds to a background, are set independently of each other as F _{bgrd_bot_thd} and F _{bgrd_top_thd} for each of these intensity _ranges . Method according to claim 1, characterized in that the intensity ranges ( 2 ) and ( 3 ) can be determined on the histogram of the recorded image using morphological filters. Method according to claim 1, characterized in that that the histogram from which the information-relevant basis the content of the recorded image is determined from a or several areas of this image to be examined becomes. Method according to Claims 1 to 6, characterized that the images are in different wavelength ranges are recorded in one or in different dynamics, in a picture of desired dynamics are summarized, wherein the dynamics of each image obtained in the desired Dynamics is converted according to the present method and the then created images using image processing methods to a be merged together.