Subpixelgenaue Kantenortsbestimmung in digitalen Mehrkanalbildern, dargestellt am Beispiel von Sensoren mit Bayer Pattern Color Filter Array

In quality assurance the inspection of geometric features of objects is one the most common tasks. There is a great variety of measuring principles. One of these is the measurement by electro-optical sensors with corresponding image processing. Image processing, on the other hand, is used for many different tasks as well. Examples are object recognition, colour measurement, scene interpretation and many more. Measurement of geometric features is one of many applications, special optimizations are therefore rarely applied for this particular applications needs. In image processing in general, the development over the last years was towards colour images or even more than three channels, so called “multi channel images”. One of the results of the advanced popularity of colour image processing is, that today some types of three channel cameras are not more expensive than their single channel counterparts. Even though these cameras are being used in system for measurement of geometric features, the algorithms used do not take advantage of the additional channels information. There are a lot of special colour image processing algorithms existing today, but there are very little concepts that address the application of measurement of geometries. In this thesis new approaches are being discussed to use the information delivered by colour image sensors in a way that the measurement of geometries in the image will be improved. Four different aspects of the chain of image processing will be addressed in this work. Two of them are applicable for all kinds of multi channel images and two are dedicated to special properties of the single most common colour image sensor type, the senor with attached colour filter array (CFA) with an arrangement according to B.E. Bayer. The two general multi channel approaches are: – Extraction of object edge information by means of a new image filter where the information of all available channels is used – High precision edge probing for those new filtered edge images with the aim of subpixel accurate edge position determination The two CFA-Sensor related aspects are: – A new “Demosaicing” algorithm to reconstruct the three channel image from the sensors raw data with special importance to geometrically correct edge reproduction – Choice for object illumination source where the interaction of the emission spectra of the source and spectral sensitivity of the senor is optimized to the needs of the designated application The new approaches presented in this thesis deliver a contribution to image processing for measurement of geometric features with multi channel images, i.e. colour images. With them, better results, respectively lower measurement uncertainty, can be achieved. While they are applicable in their presented state, they do not stand as completed system. They are meant as a new way, a concept, to utilise multi channel image data to enhance current measuring machines. In addition these concepts open up prospects to further improvement.