Star image matching by using geometric invariants

To match rigidly star images with multi-screens, a shape context based registration method by using geometric invariants was proposed. First, denoise methods, segmented method and morphology method were applied to star images to extract star points, so that centroid of star points could be calculated. Then, feature descriptors for every star point were calculated based on geometric invariants among all stars. Each descriptor was used to represent the relationship between this star and others. A price function to evaluate the similarity between star points was constructed to get a star match pair set between the two images. Finally, random sample consensus robust method was used to remove wrong star match pairs and calculate registration parameters for the space transform. Experimental results demonstrate that proposed method can successfully match the image with different number of feature points and different brightnesses in a sub-pixel level. For two group images with the size of 4608*3072 and different ISOs, the proposed method can offer the registration accuracy near 0.5 pixel, which meets the requirements of the registration.