Image Processing Algorithms — from low level to deep learning
Image processing algorithms play an increasingly important role in vision systems. The more complex a vision task, the more important is a careful selection of the best possible algorithms. The first European Machine Vision Forum aims at providing clear guidance in the world of algorithms. Developers from industry and academic researchers get the opportunity to meet and exchange their experience.
Among others, the following topics will be covered:
- How can I rate the accuracy, robustness, speed, storage requirements, and limits of an algorithm? Which metrics are to be used? Are there ways to propagate statistical uncertainty and systematic deviations from image sensors to the output of algorithms?
- How can test sequences be generated to prove the accuracy of algorithms and rate their performance?
- What are the best algorithms to detect edges, lines, and corners?
- What are the best low-level features to be used for object classification?
- How can classes of features be generated that show certain invariances?
- Modern algorithms for texture analysis
- What is the state-of-the-art of motion estimation from image sequences?
- How to match 3-D objects in 2-D images and how to detect their position and orientation?
- How do state-of-the-art algorithms work for 3-D reconstruction and simultaneous localization and mapping (SLAM)
- What are the best approaches for face recognition?
- How to detect abnormal behavior in video sequences?
- Algorithms for classification: support vector machines versus random forests versus deep learning: when does what work best?