An invariance transformation technique for interpreting images obtained from unknown operational conditions

This paper will focus on developing invariant pattern recognition algorithms for a class of parametric variations that are a significant cause of image transformations-variations in image gray level that occurs as a result of inadequate control of the imaging system. Situations such as these occur in many industrial applications-the one discussed in this paper is the magnetic imaging of gas pipeline faults. A general invariance transformation algorithm is developed and successful applications of the procedure are presented for the following two cases. The algorithm is first applied towards compensating for gray level variations in experimental signals obtained from gas pipeline inspections. The technique is then exercised with synthetic images to determine its ability to compensate for the effects of a “classical” image transformation-image scaling that occurs as a result of camera-object relative position. The results demonstrate that this invariance transformation technique can be applied effectively towards both types of image transformations