Clustering validity based image segmentation for IC water defects recognition

The authors describe an application of a well defined clustering criterion to color image segmentation for recognition of defects in integrated circuit (IC) wafers. The features of IC wafers are inherently colorful because of the interference effects taking place on the thin films which make up the IC structures. Certain classes of IC defects can be detected by the use of colors which are otherwise not possible to detect in grey-scale image processing. A typical image processing application consists of three stages. (1) image acquisition and digitization and (2) processing and quantitatively determining features in an image and (3) making decisions based on the available features. Hence, an image is first digitized and the light intensities in the image are quantized to a finite number of gray levels or RGB, three primitive colors for color image which are entered into a digital computer. The digitized and quantized image is then enhanced to reduce noised degradation. In the next stage, pixels with similar properties are grouped to form regions or segments in an image. This is the image segmentation stage.<>