Welding Wires Centerline Detection Method Based on Image Gradient Segmentation

Abstract

One of the key problems in integrated circuit (IC) manufacturing is defect detection of welding wires. In welding wire defect detection, center line extraction is a challenging problem due to the large variance of intensity value along a welding wire as against its background. In this paper, a steger centerline detection technique based on gradient amplitude is proposed for automatic extracting centerlines of welding wires. First, the image of an IC chip with a large length-to-side ratio welding wires is taken using dark field imaging method which is suitable for high dynamic reflectivity objects. Then, contrast stretching and gradient threshold techniques are sequentially used to deal with the problem of greatly varying intensity values along welding wire, which is potentially caused by changing normal vectors of the welding wire. Finally, steger center line extraction method is applied. Primary experimental results indicated that the proposed method is superior to traditional methods including threshold segmentation, maximum entropy threshold, and K-means clustering analysis in terms of conserving connectivity of extracted center lines in challenging situations with largely varying contrast of welding wires.