This paper describes an entirely new approach to low level image processing, specifically, edge detection (one dimensional feature detection), ``corner'' detection (two dimensional feature detection, including, therefore, corners, junctions, etc.) and structure preserving noise reduction. The new approach represents a significant departure from feature extraction and noise reduction methods previously developed.
The paper begins with an explanation of the SUSAN feature detection principle, and continues with details of the applications of it, including reviews of relevant past research and results of testing the applications. Because the research is based on fundamentally non-linear filtering, the approach to theoretical justification is necessarily different from that traditionally applied, for example, to ``optimal'' edge filtering.