Abstract
Spot segmentation is a crucial step in microarray image processing. A reliable and robust quantification of spot intensity can help to better identify the important genes and to reflect the true information. Most of the widely adopted methods do not perform well in terms of the reproducibility across replicates and the huge variation in the background. We propose a robust method to estimate the foreground and background intensities based on kernel density estimation without specifically segment the foreground and background pixels. We will discuss our method of gridding and expression summary for cDNA microarray images with an experiment of technical replicates. Our method is compared to K-means and Model-Based Clustering and has better accuracy across replicates.