A novel character line segmentation method for degraded binary images based on feature clustering is proposed in this paper. The application of this research is to segment character lines on images of IC chip surfaces. First, several cutting lines are detected and every two neighbor cutting lines define a candidate character line (CCL). Second, the feature of each CCL is extracted and clustering is used to choose real character lines (RCL) from CCLs. Third, a post- processing step is applied to confirm or refine the character line segmentation results. Experiments have demonstrated that this novel method can find all the character lines in degraded document images of IC chip surfaces quickly and accurately and is robust to images with background noise and logo.