|
|
|
| Statistical Properties of Natural Images | ||
| A "prior" probability model for visual images (i.e., a specification of the probability that any given image would be seen) provides a powerful constraint for many applications in image processing, computer vision, and graphics. In addition, it seems likely that the statistical properties of images are fundamental in shaping the design of biological visual systems through evolution, development, learning and adaptation. We've studied statistical properties of visual images, developed developed parametric models for these properties, and used these models in a variety of applications, and as a basis for understanding the representations used by mammalian visual systems. | ||
| Optimal inference and perception | |
| Visual perception and visually-guided behaviors are built on partial and noisy measurements. In a biological system, they are also represented with noisy elements. We've studied problems in statistical inference, in abstract form, for machine vision, and in the cotext of human vision. |
| Visual motion and Prediction | |
| When we move, the visual images projected onto our retinae change accordingly. For both biological and computer vision systems, the pattern of image velocities (sometimes called optic flow) carries important environmental information. We've studied this problem from a variety of different angles, emphasizing the issues and constraints that are common to all: |
| Visual Representation | ||
| The features that occur in visual images are often oriented (e.g., contours), and often have a particular size or scale. A wide variety of multi-scale, multi-orientation representations have been developed over the past few decades, and have proven to be important for solving problems in image processing and computer vision: | ||
| Experimental methods for fitting and testing models | ||
| We've developed methods for fitting models to physiological data, and for generating stimuli for use in perceptual or physiological experiments that are optimized for probing or differentiating the predictions of models. | ||
| Updated: October 06 2024. Created: Feb 2002. | top | |