|Ph.D Student||Topel Leah|
|Subject||Multi-Scale Processing of Digital Images for Roads|
|Department||Department of Civil and Environmental Engineering||Supervisors||Professor Maxim Shoshany|
|Professor Emeritus Yerach Doytsher|
Road recognition had been heavily dealt with in recent years. However, in spite of the intense pursuit, an efficient automatic system which is able to treat the varied widths of roads is still nonexistent.
The objective of this research is to generate automatic tools for road detection in digital images based on techniques derived from the Scale-Space Theory. Scale-Space Theory is involved with the multi-scale representation of image objects. The described method enables to analyze a wide range of road appearances in different scales without using prior knowledge.
The research hypotheses were:
· Existence of characteristic behavior of objects in different scales - This hypothesis deals with changes in the level of individual pixels, namely the inner scale.
· Objects in image exists as meaningful entities only over a limited range of scales - This hypothesis deals with changes in the general image, namely the outer scale.
On the base of these principles a system for objects detection was constructed. This system works automatically with no need for context or external knowledge. Both of these hypotheses were examined, separately and jointly, on a wide range of road images: from simple rural regions to complex urban areas.
Application of two algorithms based on each of these hypotheses separately was not satisfactory while an algorithm combining both hypotheses has yielded results with high accuracy and reliability in road recognition.