My reseach domain is computer vision and content based multimedia retrieval systems.
I am currently researching object detection and localization in large image databases. In particular, I'm trying to develop algorithms
that allow queries in image regions for object identification.
But, I am also interested of other actual reseach topics, such as:
- video/image processing, indexing and analysis,
- relevance feedback in information retrieval,
- computer vision based human-computer interaction,
- software engineering and database arhitectures.
I am very motivated by my work and 12 of my papers were accepted for national and international conferences.
In most of my papers I have addressed the issue of content-based image retrieval and I have proposed a hierarchical clustering based
relevance feedback approach. It has the main advantage of performing on the initial set of retrieved images, instead of performing
additional queries, as most approaches do. Also, I have proposed a new relevance feedback algorithm: Modified Feature Relevance Estimation.
This contains a combination between the classical Rocchio algorithm with the Feature Relevance Estimation method.
Other papers have presented the improvements of relevance feedback algorithms under the large video databases
and specialized medical databases.
In this year I have participated on MediaEval 2012 Competition, part of ARF Team. MediaEval is a benchmarking initiative
dedicated to evaluating new algorithms for multimedia access and retrieval. I was involve on two tasks:
Violent Scenes Detection Task and Tagging Task.
The Violent Scenes Detection task requires participants to deploy multimodal features to automatically detect of portions of
movies containing violent material. The second task required participants to automatically assign Genre labels to Internet
videos using features derived from speech, audio, visual content or associated textual or social information.
The databased was composed on 15.000 movies, from the blip.tv platform, and it mainly consists in web media content, rather
than classical videos, such as documentaries or TV news/programs. The main challenge of this task was the high number
of genres, e.g. for our scenario (up to 26). Also, each genre category has a high variety of video materials which may
interfere with any training step
