Multi-scale maximally stable extremal regions for object recognition

To solve the problem that maximally stable extremal regions (MSER) will become unstable when the image is blurred due to the change of scale, a novel affine invariant feature called Multi-Scale Maximally Stable Extremal Region (MMSER) which is maximally stable both in the image space and the scale space is proposed by defining a criterion to evaluate the stability of extremal regions in scale space. And a fast extraction algorithm based on N-tree disjoint set forest and seeded growing algorithm is designed for MMSER. At the same time, according to the property that MMSER can describe the contour of local features fairly well, a new kind of descriptor is designed for MMSER by combining the local gray grads and the shape context information. The experimental results prove that MMSER is much more stable and discernable under different affine transformation conditions.