Application examples and methods

Task#

In automatic visual inspection and image processing, individual views are often not sufficient for the evaluation of scenes and the quality inspection of objects. Examples are complex shaped objects where shadows prevent suitable illumination and impair the observability of the entire surface. Also for other tasks, important geometrical and optical properties of the object or scene cannot be captured with single images.

 

Fig. 1: Variation of the focus position: The indentation cannot be shown completely in focus in a single image. Fig. 2: Variation of the azimuth of a directional illumination: The scoring structure is only visible when the illumination falls perpendicular to the scoring.

Figure 1 shows a depression whose extension is greater than the depth of field achievable with the imaging optics. If the focus position is changed, every single location of the object can be imaged sharply, but not the entire image section in a single exposure.

Even if an image with suitable acquisition parameters would be sufficient in principle, it is often difficult to identify and adjust the optimum parameters. As an example, Figure 2 shows a groove texture where the groove structure is only clearly visible when the illumination falls perpendicular to the grooves. In the presented image series with varied azimuth of a directional illumination this is the case twice per revolution of the azimuth, whereas with an illumination in the direction of the grooves almost no groove structures are visible.

Solutions

In such cases, a promising approach is to acquire an image series instead of a single image, in which certain relevant parameters of the image acquisition are specifically varied. All parameters of the image acquisition that are accessible to technical influence can be varied. From such an image series a result image can be obtained by concentrating the information distributed over the series by means of a fusion.

This strategy corresponds to the approach of a human observer, who proceeds similarly during visual inspection: An object is examined under different perspectives and with different illumination until the desired property is determined.

 

Results

The image series shown above were examined for characteristic features that vary in the image series. These characteristics were classified in a suitable way, so that e.g. by evaluating a local contrast measure a synthetic focus extension is possible.

 

Fig. 3: Fusion result of the focus series of picture 1: Due to the synthetic extension of the focus area, the entire deepening is sharply illustrated.
Fig. 4: Fusion result of the illumination series of image 2: Segmentation of the scoring area, the background is masked in black.

Methods#

System theory

For the acquisition of image series a system theory is developed, which represents the theoretical basis for the meaningful acquisition of image series. Special emphasis is put on the representation of the available parameter spaces and their mutual dependencies.

 

Variable acquisition techniques

Acquisition parameters are identified, the variation of which enables the extraction of additional information. For the acquisition of image series, techniques are developed to adjust the parameters to be varied with the help of a computer. Emphasis is put on the handling of the components of image acquisition, the development of suitable illumination devices as well as the development of illumination strategies and their experimental implementation.

 

 

Fusion of image series

The information of interest is extracted from image series by fusion. Different fusion processes are theoretically analyzed and further developed for application to image series with multivariate image constellations.

Active Vision

From the evaluation of single images and the fusion results of an image series, the suitability of the acquisition parameters for the object or scene under investigation can be concluded. By optimizing the acquisition parameters in consideration of previous results, a kind of control loop is developed, which leads to an optimal set of parameters for the object or scene. The static "looking at" of the object usually practiced in automatic visual inspection is thus to be replaced by a dynamic "looking at" that imitates the visual inspection of an expert.