C. Krger, C.M. Bartle, J.G. West (New Zealand), and V.-H. Tran (USA)
Image Analysis, Object Detection, Segmentation, Threshholding
We are presenting a technique to analyze digital x-ray images of industrial products to detect undesired foreign objects. Products are made of fairly homogenous material such as meat in a cardboard container. The main advantage of the method is its high speed and low false detection rate. Image retrieval is performed through x-ray scanning of the entire container, where the attenuated signal is recorded at the detector arrays. Spectral analysis of the resulting intensity image will filter intensities not belonging to the main content of the container, assuming that the main packing follows a distinct intensity distribution. This distinct intensity distribution is determined for each image as a base for further analysis. Filtered suspected intensities can be used for the rendering of the object or to notify detection for further action. Several advantages follow from this technique: Firstly, it is independent of the shape of the object as no edge detection is needed. Secondly, the calculations can be performed very fast, as only part of the intensity spectrum is used for the analysis, and thirdly, very little false detections are encountered. For these reasons the technique is ideal for inline purposes such as conveyor belt driven applications.
Important Links:
Go Back
