High speed, multi-scale tracing of curvilinear features with automated scale selection and enhanced orientation computation

We propose a new high speed line tracing algorithm based on a well known differential geometric line extraction algorithm. The previously separate steps of line detection and line tracing are performed simultaneously. This allows the exclusion of non-candidates from processing. Exploiting the inherent continuity of lines and using extracted line characteristics in subsequent detection/tracing also solves the problem of multiple, computationally expensive scale space iterations. Consequently, processing time is shown to be reduced by up to a factor of fifty. Furthermore, the extraction is very sensitive as hard to set global thresholds are no longer required. In the context of these proposals, we also review methods to identify the pixel-wise line orientation. The previously used orientation of maximum second derivative proved to suffer from systematic errors, whereas, our two novel methods proved more reliable. Our algorithm is designed for images containing only a single line but can be applied to images with multiple lines, especially if the global image structure is known.