6533b82ffe1ef96bd129626d

RESEARCH PRODUCT

DEFINITION ET ETUDE DES PERFORMANCES D'UN DETECTEUR DE STRUCTURES CURVILINEAIRES. APPLICATION A LA STEREOSCOPIE ET LA DETECTION D'OBJETS FILAIRES

subject

description

The work presented in this paper falls under the domain of matching. In this context, we try to show the interest of using the curvilinear regions in order to enhance the global performance of various matching applications. A brand new detector for curvilinear regions is introduced (CRD), along with the specific curvilinear regions descriptor (CRD) required. Performance evaluations were done in two parts: First, we evaluated and compared the segmentations performances of our detector, in particular on a retina vessels segmentation application. And second, we estimated the matching performances using K. Mykolajczyk toolkit. This methodology helps obtaining intrinsic performances and comparing them to a so-called "standard detectors" We prove our CRD robustness against multiple disruptions e.g. viewpoint change, scaling, luminance factor, JPEG compression effect and blurring effect. Moreover we prove that it can complement the standard detectors with a wide range of considerable information. The presented applications can be summed into two types. At first, we present two segmentation applications: Glue line detection on pre-assembled motor cylinder head, as well as the aorta detection in a sagittal cut using MRI images. The second application type concerns matching operations. For this class of application, we propose an automatic homography estimation.