Intelligent methods for Image Processing and Analysis

Researchers

  • Sven Loncaric, principal investigator
  • Damir Sersic
  • Marko Subasic
  • Tomislav Petkovic
  • Hrvoje Kalinic
  • Vedrana Balicevic
  • Hrvoje Bogunovic

Institutions

  • Faculty of Electrical Engineering and Computing, University of Zagreb, Croatia
  • Faculty of Medicine, University of Zagreb, Croatia
  • Faculty of Geodesy, University of Zagreb, Croatia

Project duration

The project duration was from 2007-2012.

Funding agency

Ministry of Science, Education and Sports, Republic of Croatia.

Short description

Image analysis and scene interpretation are complex tasks that require knowledge about objects contained in the scene and about their mutual relationships. Conventional image analysis approaches are based on relatively simple techniques that only utilize low-level information obtained from pixel intensity values. The major limitation of these approaches is lack of high-level knowledge.

The goal of the project is to develop intelligent knowledge-based methods for object detection, recognition, and tracking that are robust and accurate. The proposed approach utilizes both low- and high-level knowledge for scene interpretation. To achieve this goal we will employ knowledge-based techniques such as 3-D model-based methods, neural networks, evolutionary algorithms, expert systems, and intelligent agents and apply developed techniques to the problems of medical, face, and range image analysis.

The hypotheses of the project are:

  1. High-level knowledge in the form of object models and scene composition models combined with low-level intensity-based feature extraction improves accuracy and robustness of object recognition and tracking.

  2. Fusion of information obtained from multiple imaging modalities (range image data and photographic data) improves accuracy and robustness of object recognition.

We will develop new methods for detection, recognition, and tracking of objects from single and multiple 2-D and 3-D images. We will develop methods for:

  1. 3-D vessel tree segmentation from computed tomography images,

  2. Segmentation and 2-D/3-D tracking of a catheter inserted into the blood vessel,

  3. Knowledge-based approach for face image analysis,

  4. Fusion and analysis of range and photographic images.

The proposed methodology will be tested on 3-D and 2-D medical images obtained by computed tomography and 2-D X-ray imaging for catheter tracking. To verify face image analysis methods, we will use FERET face image database. LIDAR range images and corresponding photographic images of a terrain will be used for verification of range image analysis methods.

Development of intelligent image analysis methods has great importance for realization of computer vision systems. Computer vision systems have applications in many areas of human activity. 

Publications

The project has resulted in a number of publications that can be seen here.

Back