Signal Processing and Communications
P.O. Box 70 08 22
Aim is the segmentation and classification of images in the maritime environment. For this purpose, mainly convolutional neural nets are used. Novel methods are investigated and, if necessary, adapted or developed. Classical image processing algorithms are used to support segmentation.
To understand the scene of the image, new interesting classes like water, sky, … and many more are learned.
To support the classification task, a contour extraction method with deep learning was implemented.
The RefineContourNet uses an established segmentation network and reaches State-of-the-Art in edge-detection on the BSDS500 Benchmark: BSDS500 SoA
Test-Code on https://github.com/AndreKelm/RefineContourNet
Supervised project work
Vijesh Sao Rao – Training State-of-the-Art CNNs for Segmentation and Contour Detection on Ship Images, Student Project
Christina Sander – SSD Analysis and Development of Adaption Proposal for a CNN Ship Detector, Student Project
Supradeep Chikaballapur Manjunath – Contour Detection using SRCNN, Student Project
Karthik Kadur Manjunath – Analysing Hypercolumn Features for Object Contour detection and Semantic Segmentation, Student Project
Supervised Master Theses
Vijesh Sao Rao – Generating synthetic NIR images from RGB images using GANs, Master Thesis
Rocío Aldana Figueroa – CNN for Detailed Ship Image Segmentation using Contour and Segmentation Feature Maps, Master Thesis
Sous-Lieutenant Jean Fissot – Using Phase-Stretch-Transform Algorithm as an Image-Feature-Extractor, Master Thesis
A. Kelm, V. Rao, U. Zölzer: Object Contour and Edge Detection with RefineContourNet – Computer Analysis of Images and Patterns, 18th International Conference, CAIP 2019, Salerno, Italy, September 3–5, 2019, Proceedings, Part I
P. Bhattacharya, G. Simkus, C. de Obaldía, A. Kelm, U. Zölzer: Convolutional Neural Networks for Digital Signal Processing, LSA2017 – Lübeck Summer Academy on Medical Technology, July 2017, Lübeck, Germany.
Letzte Änderung: 30. March 2020