Adaptive Multi-Threshold Object Selection in Digital Images

Abstract

A new algorithm for adaptive selection of compact and extended objects is investigated. The algorithm is based on the initial multi-threshold processing of the original monochrome image, which creates a set of binary layers. On their basis, using the percolation effect, a three-dimensional hierarchical structure is constructed that allows solving the optimization problem, i.e. choosing the best binary layer for each object in terms of the geometric criterion used. The key idea of the algorithm is that the solution is based on a posteriori information about the properties of objects that can be selected from each binary layer. Using this information, you can successfully solve adaptive selection problems, while maintaining the shape of each object of interest, despite the nonstationary background. In the test problem of detection against the background of Gaussian noise, the use of selection provides a gain in the signal-to-noise ratio of at least 6 dB. The results of selection of objects on typical noisy model and real television image show the efficiency and effectiveness of selection of compact (spotted) and elongated objects of interest with minimal distortion of their borders at a fairly low signal-to-noise ratio.