The paper describes a new stroke extraction approach to identify the feature points of a character, using line-filtering and learning-based techniques. The line-filtering technique based on the convolution operations with a set of l-D Gabor templates is efficient in extracting the stroke segments of the character and robust in noise tolerance. Furthermore, unlike convectional feature-point detection techniques where decision rules and thresholds have to be specified, our learning-based technique for feature-point identification implicitly represents the rules and thresholds without further parameter adjustments. Experimental results show that the learning-based technique is capable of generalizing the learning knowledge to identify feature points and can get an average identification rate of 95.27% for hand-printed test characters and 96.78% for machine-printed test characters.