Annotation of images by intelligent image processing

OData support
Dr. Szűcs Gábor
Department of Telecommunications and Media Informatics

Nowadays the size of digital data and database are constantly growing and developing. Basic part of these are digital pictures which are used by users on a daily basis thanks to digital photo shoot and photo sharing. Significant part of this bunch of accumulated pictures are the ones taken in nature (outside) called nature photos. Classifying and processing of these pictures are not easy and not a solved problem.

In my task I have designed and developed a system which is used for processing nature photos. My system makes it possible to follow up the changing of flora and fauna which was captured on photos without being on sight by any specialist. The way this is realised that the user takes picture of the nature which will be given access to the system. First step of processing photos is segmentation so objects become separated from the background. As a result every single processed picture is represented by a feature vector which is produced by feature extraction. These preprocessed images are clustered via clustering methods then the pictures with the same content are classified and annotated. As a result of this, pictures with same objects are sorted to the same group therefore classifying and following up on their changes are much easier and can be automated.

In my thesis I present the system I have designed and carried out. I demonstrate particular subsystems and detailed, mathematical matters. I describe object localization, object detection and the steps of segmentation also the different and specified feature extraction techniques as well as the applied clustering procedures on the produced vectors.

In the last part of my essay I present a detailed evaluation of the operating system. In doing so I demonstrate the different parametric options and inspect that in different circumstances via which parameterization can the best result be achieved.


Please sign in to download the files of this thesis.