Thanh-Hai TRAN
HomeReasearchPublicationsTeachingLinks


Ridge Extraction

1. Why ridges ?  In order to represent objects for recognition, many kinds of feature have been studied: region, contour line, skeleton, ect. Recently, peoples talk alot of interest points (sometimes called natural points). Interest points have all properties that we wait for to be a good feature as accurate localisation, invariance to illumination and scale. Recognition and matching based on interest points obtain impressive results. However, one problem is interest points are not always the best features for representing images, mostly images containing oblong structures such as veines in medical image, roads in geographical images. To have an explicite representation about structures in images, interest lines (ridges) are the most convenient. As we can see below, ridges studied at different scales represent global shape as well as details of the object.

2. What is a ridge: Intuitively, a ridge is to be though of as the path you follow on a mountain, where there's always a drop both to your left and to your right. In an image, a ridge occurs when there is a connected sequence of pixels having intensity values which are higher(lower) in the sequence than those neighbouring the sequence. With this intuitive definition, a ridge can be considered as an approximation medial axis of an oblong object such as a road in a satellite image, a blood vein in a medical image, etc. 

2. How to detect ridges in image ? When we apply the Laplacian operator on the gaussian of the image, ridge points are extrema points. Hence to detect ridges, we compute Laplacian of gaussian of the image and localize its extrema. For all detail, see my thesis.

 

2. Ridges at multi-resolutions: An object can be represented by ridges detected at several scales. At small scale, ridges represent details of the objet, at bigger scale, ridges represent global structure. Below shows two examples in which ridges at one scale cannot represent all structures of an object.  

 

 

3. Examples of ridges in natural images

Ridges represent lines on the zebra

 

Ridges reprensent skeletons of fingers

 

Ridges (blue lines) represent sleletons of the leaf while interest points (red points) reprensent its coins. 

 

Representation of  fingerprint by ridges (red lines) and valleys (blue lines)

4. Example of ridges detected at several scales

5. Applications
- Text detection
- Human modelisation