Extraction of small target based on local extreme convergence

Automatic detection of small targets in the complex context is still not perfect, an algorithm was proposed that used all paths converging to the same limit point to describe the small target area. The starting points of the path were screened based on the image gradient features. A path starting from the starting point along the gradient direction of steepest descent converged to a local minimum point, and all the paths that converge in the same path constituted an independent core region. The difference was analyzed in the target features between the focus independent core region and noise independent region, and the gray average ratio of the target features inside and outside was used for the independent core area filtering, the focus of the core area was obtained. The region of the target core was obtained by polymerizing the independent core region. The experiments show that the algorithm can automatically detect the focus target, and compared with existing algorithms, it increases the degree of automation of the small object extraction, have a strong robustness.