Research

Introduction

The research of the Images and Visual Representation Lab (IVRL) in the School of Computer and Communication Sciences (IC) is primarily concerned with color image processing and computer vision for computational photography and video to optimize the quality of natural image encoding and display. The developed imaging systems, algorithms, and models are based on low and high level processing features of the human visual system as well as the underlying physics of the acquired and displayed light field.

In general, we are interested in the question of “what is an image?” What are the fundamental components that “make” an image, and how does the change of one of those components influence the perception of it? How come we always know if we are looking at a “good” image or a “bad” image, even without being able to identify what is “good” or “bad” about it? Can we identify models and algorithms that define and predict these subjective criteria, and build systems that improve the visual experience? In the domains of photography and video, where we only acquire a reduced dimension of the original light field, there are still fundamental gaps in our knowledge of how much other concurrent physical stimuli as well as mental processes contribute to perception.

Similarly, it is very easy for humans to “understand” an image and to extract meaningful information from it. Yet, machine vision systems still have difficulties to perform the same tasks, such as recognizing and tracking an object, labeling regions, and categorizing scenes. We investigate which low and high level human vision properties can be translated to algorithms that simplify these tasks, and what additional physical information might be useful to improve current performance.

To develop meaningful computational photography and image quality models, systems and algorithms that better mimic visual perception needs the collaboration of several fields, including computer vision, image processing, machine learning, psychology, physiology, and physics. IVRL is always collaborating with experts from different fields and different institutions.

Applications of the developed systems, methods and algorithms are in new camera technologies, cross-comparisons of different image capture, processing and display techniques, in image understanding, and in digital image production, printing, and archiving. For example, IVRL has developed demosaicing and high dynamic range image rendering algorithms that are based on a retinal processing model, new camera models that use silicon sensors’ inherent sensitivity to near-infrared to improve image quality and image understanding, automatic image enhancement and visualization tools that take into account the specific context of a photograph, low-level saliency detection that aid several applications, and superpixels that facilitate image segmentation.

Research Support

IVRL’s research is supported by the Swiss National Science Foundation and the Hasler Foundation.
IVRL collaborates with OCE Print Logic Technologies (a Canon Group).
 

IVRL was a member of the National Competence Center in Research on Mobile Information and Communication Systems (NCCR-MICS), a center supported by the Swiss National Science Foundation under grant number 5005-67322 (2001-2012). IVRL was also supported by the Xerox Foundation.