Skip to content
Research Fellows Directory

Carlos Thomaz

Professor Carlos Thomaz

Research Fellow


Imperial College London

Research summary

Faces are familiar objects that can be easily perceived and recognized by humans. However, the computational modelling of such apparently natural and heritable human ability remains challenging. To understand and emulate how humans accomplish the process of coding and decoding high dimensional visual patterns that may be metrically very close to each other, such as facial images, it seems necessary to study dimensionality reduction methods that enable a selective treatment of the attributes that compose the patterns of interest using some prior knowledge about the problem under investigation. The main objective of this current research is to develop statistical pattern recognition methods that incorporate human reasoning as prior knowledge to address our distinct and selective perception of facial patterns. More specifically, we aim to investigate the interplay between low-level visual attributes, such as texture represented by intensities of pixels, and high level visual attributes, represented by semantic concepts of human reasoning, to extracting and interpreting the most discriminant features in face image analysis. The high level visual attributes are described by supervised information like gender and facial expression, available on training samples and quantified by either statistical significant differences explicitly calculated from the data or cognitive relevant associations expressed implicitly by human visual perception. We expect as result of this research the development of statistical pattern recognition methods that can be used not only as pre-processing step for automated recognition systems, but also as a conceptual framework for human reasoning and coding of face images.

Grants awarded

Learning and extracting priori-driven representations of face images to understand the human visual recognition system

Scheme: Newton Advanced Fellowship

Dates: Mar 2015 - Feb 2018

Value: £35,000