3D Video: From Capture to Diffusion

Chapter 6

Feature Points Detection and Image Matching

6.1. Introduction

Image matching is a part of many computer vision or image processing applications, such as object recognition, registration, panoramic images and image mosaics, three-dimensional (3D) reconstruction and modeling, stereovision or even indexing and searching for images via content. The problem is finding the geometric transformation (rigid, affine, homographic or projective) that best matches two images using visual information shared by the two images. The main hypothesis is that the visible part of the 3D object from a given angle is almost identical to that obtained from another angle.

The different matching techniques vary according to the number and the nature of visual information used. When all the pixels in the image are used, matching is dense. Within the context of medical image registration, this may relate to landmarks, precisely defined anatomical points [ROM 02], or “semi-landmarks” [BER 04]: the amount of visual information used is minimal but matching is preceded by a landmark recognition stage. This chapter will examine a general scenario where the amount of visual information is limited and where no prior knowledge is available.

This visual information corresponds to regions detected using specific photometric or geometric properties such as edges, corners or “blobs”. They are represented by local descriptors often based on local photometry. Matching visual elements shared by two images is carried ...