Introduction to Image Analysis and Vision Research

Image Analysis involves creating a mapping from a digital image to a description of the image's content. That description of content might be a number (there are 4 epithelial cells and 2 squamous cells in this microscope image), a physical measurement (the volume of a chamber of the heart), a degree of anomaly (shape variation from normal caused by pathology), a labeling of the image's pixels (turn the kidney green and the liver brown), or a 2-D or 3-D representation of the image content suitable for input to a computer graphics renderer.

Our main objective is to understand fundamental properties, principles, and processes of visual perception. How do we perceive objects? How do we represent an object so that we can recognize it later? To answer these questions, we develop and use mathematical and statistical tools that describe visual shape. We also test and sharpen our mathematical models by using our best methods to solve real problems defined by researchers in a wide range of disciplines that are concerned with images. Our research then contributes to science both by solving a real problem of interest to our collaborators and by advancing our understanding of visual perception.

Most of our image analysis work involves biomedical applications, but we have also worked with image analysis problems from astronomy, crystallography, earth science, and text analysis. Biomedical problems are particularly attractive to us because they usually involve relatively small images, there is available expertise to tell us whether we have really solved the problem, and good solutions to these problems can really improve human lives by improving the ability of physicians to diagnose and treat diseases. The immediacy of the need for solutions to biomedical image analysis problems gives our laboratory its sense of drive and urgency.

Vision research attempts to understand the capabilities and limitations of the human visual system. (Other biological visual systems are of interest to us, but we do not perform that research in our group.)

We perform experiments in visual psychophysics that test how well the performance of human observers is captured by our mathematical or computational models. Some researachers in our group are interested in testing whether different computer-based display mechanisms (large video screens, head-mounted displays, high definition screens) perform as effectively as conventional display methods such as X-ray images on film or CT images from standard viewpoints. Others are interested in devising computational mechanisms that simulate properties or capabilities we observe in human vision.