In real-time video stream application, a feature point detector is very necessary. Stable feature points are useful in machine vision application. Feature point detectors such as SIFT, Harris and SUSAN are good methods which extracted high quality features, but neither the Harris detector nor the detection stage of SIFT can operate at real-time frame rate. In this paper, we present a new corner detection algorithm which uses fuzzy similarity matrix to calculate average similarity of contiguous pixels. It can be used in real-time applications of any complexity. In our experiments to evaluate our corner detector, difference image transformations are computed and the repeatability of feature points between a reference image and each of the transformed images are computed. Experiment results show our corner detector could give good results.