Covariant support region and detection algorithm based on LoG corners

Detection of local feature covariant region is a new technology of image contents and image semantic representations, and it has become an important foundation of the image recognition, learning and understanding. First, a Laplace of Gaussian corner detection method is proposed based on edge contour curves, in the meantime, a new local feature descriptor, named covariant support region, is introduced. Then, a detection algorithm of covariant support region is framed, which is covariant for rotation and scale transformation. Comparing with previous studies, the computational complexity of proposed algorithm is significantly reduced by this method. The experiments data indicate that the method proposed in this paper has good performance on higher accuracy, higher repeatability, and lower complexity.