NEURAL NETWORK BASED DETECTION OF HETEROGENEITIES IN NOISY IMAGES
Sergey Abramov, A. Naumenko, Владимир Васильевич Лукин, Sergey Krivenko, Igor V. Kaluzhynov · Telecommunications and Radio Engineering · 2020
Many methods of image processing include a stage of detecting heterogeneities (edges, small sizes objects, textures). It is often difficult to reach efficient detection due to noise presence in analyzed images when conventional detectors fail. Neural networks are tools that allow to partly improve detectability of heterogeneities due to joint use (aggregation) of elementary detectors. Performance can be improved due to proper selection of elementary detectors as well as pre-processing (pre-filtering) or post-processing (aggregation of detection results). In this paper, we consider some of aforementioned aspects and give examples of neural network learning and application to different test and real life images.