IMAGE SEGMENTATION AND OBJECT SELECTION BASED ON MULTI-THRESHOLD PROCESSING

Abstract

Introduction. In order to automate data processing in remote observation systems using television and infrared cameras, synthetic aperture panoramic radars, as well as laser and acoustic systems, it is essential to be able to reliably detect, isolate, select and localise objects of various shapes in images. Objective. The development of a methodology based on multi-threshold analysis. Materials and methods. The developed image segmentation and object selection approach having optimal selection threshold assessment is based on the results of multi-threshold image analysis. Results. Based on the analysis of a series of standard objects with known shapes hindered by synthetic noise, as well as representative examples of remotely sensed images of the Earth’s surface, improvements in the characteristics of both entire image segmentation and selection of particular objects according to several objective criteria were achieved.Conclusion. The main advantage of the proposed approach consists in the minimisation of the post-processing shape modification of the selected objects. Although this is achieved at the cost of the resource-consuming multi-threshold analysis procedure for each processed image, this can be also partially compensated by the simplicity of the algorithm and its possible parallel implementation.

Keywords

• многопороговая обработка
• сегментация изображений
• селекция объектов
• метод бинарного интегрирования
• вероятностные модели