A Research Fellow (postdoc) position and Research Associate (Meng/MSc/BEng) position are available in the School of Computer Engineering, NTU, for an A*Star-funded project under Singapore-Poland research collaboration program. Project Title: Framework For Visual Information Retrieval And Building Content-Based Visual Search Engines.

Project Description:

The project aims to develop a flexible and adaptable framework for description and annotation of images, both stored in databases and captured online. The proposed framework would be subsequently used for:

(1)  novel organization of visual databases allowing quick search using image annotation;

(2)  visual information search and retrieval, and eventually

(3)  development of tools for building search engines based on visual contents.

The proposal results from mutually complementary researches conducted by Singaporean and Polish partners. Singapore partners contribute results in machine/real-time vision methods, in particular methods used for learning and detection of particular image contents under partial information conditions (unstructured environments, poor quality of images, occlusions, etc.), and new compositional query formulation and processing based on structured and adaptive similarity matching. Polish partners will contribute their novel techniques of image annotation and image interpretation. Image annotation methods, together with spatial relations analysis is going to be used for high-level scene description. Usage of contextual information and visual knowledge base allows to discover not directly detectable image features. Both sides would be jointly working (with a participation of world-class experts) on the optimized incorporation of AI methods at various stages of image description, annotation and retrieval. Additionally, participation of highly-qualified medical staff in the Polish team would facilitate adapting and testing the developed framework in selected areas of medical imaging, as a specialized case-study application.

The hierarchical image description/annotation starts from a rigorous mathematical model at the lowest level to detect local (and invariant) primitive features and to quantify the confidence of detections. Development of hardware accelerators for the pre-processing phase is also envisaged. At higher levels, AI mechanisms are gradually introduced to identify more complex features and to verify the presence of lower-level features, which eventually leads to the automatic image annotation symbolically representing its content at various levels of abstraction. Such image annotations become a part of the database entry containing the corresponding image and can be subsequently used to search for images of relevantly similar visual contents. A combination of AI mechanisms (e.g. approximate reasoning, neural networks, fuzzy logic) and mathematical tools (dedicated systems of equations) would be used to identify structurally matching sub-contents of images and to verify the validity of hypothesized matches to support an intuitive and flexible query interface for expressing semantic similarities, beyond current simplistic example-based and text-based methods.

The framework would use an open library of routines for detection elementary (and higher-level) structural features so that the method can be continuously expanded (e.g. by adding more features at all level, and more features-detection mechanisms) and adapted to more specialized applications. Initially, general categories of images would be studied to test the level of generality of the framework. Subsequently, selected medical databases may be used as the intended feasibility-study area.

Job Description:
The positions are available immediately. 2 year contracts with a 1 year continuation are offered. RF and RA will work closely with the Principal Investigator andr with collaborators from NTU, Wroclaw University of Technology (Poland) and I2R. RF would be a leader of the team including the RA and 1-2 PhD students. The team will work on development of low-level feature detectors and integration of the detectors with high-level image annotation schemes (including the usage of approximate reasoning/NN/etc. methods). RA will be also responsible for setup and maintenance of the project website. The candidate should ideally have some knowledge and experience in  imaging processing and/or machine vision.
As specified in the project requirements, Singaporeans, Singapore PR’s and citizens of Poland are particularly welcome.

Anyone interested please send a CV to the Principal Investigator Assoc. Prof. Andrzej Sluzek (assluzek@ntu.edu.sg).