We propose the new multiframe feature point detector for two or more images of the same scene, which are supposed to be blurred, translated and rotated with respect to each other and noisy. Feature points are here understood as salient points with high local contrast in the image, perceived as "corners". The repeatability condition - result of the method should not be affected by change of imaging geometry, radiometric conditions and by an additive noise - is fulfilled by the introduced algorithm. The detected sets of features in images have sufficiently high number of common elements due to the invariancy of the method under supposed image degradations. The performance of the method is demonstrated on satellite images and compared with other feature detection methods. Detected features form the input information for further processing like image registration and fusion, multitemporal analysis or object recognition, among others. The image registration application of the feature detector is introduced together with the experiment results (satellite imagery, computer vision).