Scene Understanding

Principal Investigator: Prof. Joni Kamarainen

Our research aims at computer vision based total scene understanding. Relevant sub-topics are object recognition, object detection, classification and visual categorisation. All these are some of the most investigated topics in computer vision, but the big theory is still missing. We develop methods which can be used in efficient visual search and indexing in large scale image and video databases and Internet.

Our main interest for the past few years have been part-based models of visual classes and statistical methods for their learning and detection. Our methods are general and can be used to learn and detect any objects or object classes. Our main scientific contribution is a novel local feature referred to as "simple Gabor feature space" or "multi-resolution Gabor feature", which provides robust and invariant feature for learning and detecting local object parts. Our main focus is now on spatial (constellation) models of local parts and on semi-supervised and completely unsupervised methods for visual object categorisation. Moreover, we put special emphasis on large scale problems of visual classification.

Featured projects

Big Visual Data

The goal of this TUT capital gain funded project is to study new methods and applications of big data. The project is multi-disciplinary and consists of six different topics. Our topic is to investigate potential of big visual data and efficient methods for processing a huge amount of visual information (images and videos). In particular, we want to understand the complex networks of millions of images and be able to detect and learn various visual classes and their attributes unsupervised or semi-unsupervised from data and metadata.

Gabor Object Detector

In this project we investigate one of the most fundamental problems in computer vision and artificial intelligence - detection, localisation and recognition of objects in images (2D and 3D) and video. A method should be very robust similar to the human visual system and easily cope with imaging distortions, occlusions, pose changes, class variation etc. We have focused on our long-term research on Gabor features and currently we have a very effective part-based generative model that is easy to train for new classes and easy to use. We further develop our methodology to compete with state-of-the-art.


The goal of this Academy of Finland funded project is to solve some of the most fundamental yet challenging problems in computer vision - object detection, localisation and recognition - in large-scale. With the development of hardware, people nowadays have much more images in the size of G bytes than before. In the light of this, the research in the direction of processing big data attracts wide public attention. Due to large-scale visual object categories, the inter-class variation and computational complexity will further increase the difficulties in addition to large intra-class variation. The main focus of this project is set to improve the accuracy and efficiency of the-state-of-arts for large-scale visual object recognition.