A defect identification algorithm for sequential and parallel computers

The comparison of images containing a single object of interest, where one of them contains the model object, is frequently used for defect identification. This is often a problem of interest to industrial applications. This paper introduces sequential and parallel versions of an algorithm that compares original (reference) and processed images in the time and frequency domains. This comparison combined with histogram data from both domains can identify differences in the images. Extracted data is also compared to database data in an attempt to pinpoint specific changes, such as rotations, translations, defects, etc. The first application considered here is recognition of an object which has been translated and/or rotated. For illustration purposes, an original image of a computer-simulated centered needle is compared to a second image of the hypodermic needle in a different position. This algorithm will determine if both images contain the same object regardless of position. The second application identifies changes (defects) in the needle regardless of position and reports the quality of the needle. This quality will be a quantitative measurement depending on error calculations in the spatial and frequency domains and comparisons to database data. Finally, the performance of sequential and parallel versions of the algorithm for a Sun SPARCstation and an experimental in-house built parallel DSP computer with eight TMS320C40 processors is included. The results show that significant speedup can be achieved through incorporation of parallel processing techniques