Object recognition techniques in real applications
Laura Fernández-Robles · University of Groningen research database (University of Groningen / Centre for Information Technology) · 2016
This thesis proposes and evaluates object description and retrieval techniques in different real applications. First, we addressed the classification of boar spermatozoa according to acrosome integrity, which is an important challenge in the veterinary field. We presented several methods based on invariant local features. We yielded satisfactory results using a concatenation of SURF and global texture descriptors and k-NN classification algorithm. Secondly, we focused on the implementation of computer vision solutions for tool wear monitoring, which is a key issue for extending lifetime of cutting tools. We provided two new methods for insert localisation and an automatic solution for the recognition of broken inserts in edge profile milling heads. The proposed approaches are efficient and can be set up in-process without delaying any machining operations. Finally, we worked within the European project Advisory System Against Sexual Exploitation of Children. One of the most challenging tasks in this project was to find specific objects using content-based image retrieval. We evaluated different clusterings of keypoints for object retrieval and proposed a new descriptor, named colour COSFIRE. Colour COSFIRE filters add colour description and improve the discrimination power to COSFIRE filters as well as provide invariance to background intensity. This thesis contributes to the understanding and provides effective solutions of real applications using object recognition and image classification techniques.