Mark Ruzon's Research Interests

My Research Interests

Mark A. Ruzon

Two of the major modes of visual perception are form and color. Form, which refers in general to any recovering of structure from an image formed by a camera sensor (including the human retina), can be discovered independently of color. This statement is trivially obvious, as we can watch black and white televisions with negligible loss of understanding. It would, therefore, seem prudent to study form in grayscale images, since it is a sufficient condition for recovering image structure.

However, it is undeniable that color adds an immeasurable quality to our lives. Color in the natural world serves four main purposes: (1) it allows us to easily interpret complex data (such as a roadmap), (2) it denotes affiliation or identification of objects or people, (3) it provides richness of visual experience (after all, who watches television in black and white anymore?), and (4) it triggers emotional responses.

It is the goal of my research to explore the relationship between color and texture, which, along with shape, is one of the primary manifestations of form. Texture is a catch-all term for a set of perceptual techniques that abstract large amounts of visual information into relatively small amounts based on "patterns" or "features" present. The type of information used is mostly photometric, meaning that we are interested in the colors and intensities of groups of pixels. While geometric information is certainly important, it is not nearly as critical to texture processing as it is to shape processing. Thus, the two modalities are distinguished.

Color and texture share only a few similarities, but they complement each other in many ways. Color and texture can both be modeled as vectors, but whereas color exists at a single pixel, a texture vector represents a large neighborhood surrounding a pixel. We can speak of discontinuities between textured regions as well as color pixels, but in color we refer to them as `edges' while in texture they are thought of as `boundaries,' emphasizing the size of the support as well as the role noise plays in detecting each.

The application of this research is to the problem of segmentation, or how to divide an image into "meaningful" regions based on properties that are shared within a region but distinct from other regions. In order to accomplish this task, a number of questions must be answered: