Endoscopic image analysis. Image mosaicing

Endoscopy enables the visualization of cavities inaccessible to the eye, using an instrument called an endoscope. The endoscope is usually introduced into the cavity to be explored through the available openings. Widely used in medicine, endoscopy is also used in industry for visual non-destructive testing. This article deals with medical endoscopy, but the general issues are the same in industrial endoscopy.

The endoscope is either a rigid tube or a bundle of optical fibers of various lengths and diameters, depending on the application. On one side of the instrument (distal end) is an optic and a light diffuser to illuminate the scene to be viewed, and on the other side (proximal end) is usually a digital image sensor and a light source. Images and light are conveyed by an optical channel between the two ends. The endoscope is often manipulated manually by an operator, who observes the images on a monitor. Defects in endoscopic images are related to the optics. Uniformity of illumination in situ is difficult to achieve, especially as the operator manipulates the instrument by hand, with variations in distance between the distal tip and the observed surface that can be significant. The focal lengths of optics are chosen to be short, so as to have large aperture angles, which introduces vignetting and geometric distortion. Organs examined in medicine can be specular in nature, generating annoying reflections on images. Finally, the optical channel for image transmission may be made up of optical fibers (flexible endoscopes), leading to a so-called "honeycomb" effect on images. This article describes in some detail the techniques that can be used to correct the above-mentioned defects in endoscopic images.

Today, for archiving purposes or patient follow-up over time, endoscopic examinations are recorded in the form of a video film, which can be of long duration. Mosaicking techniques have been developed to present the information contained in the video-sequence images in a much simpler and more compact form. These involve calculating the geometric displacements (e.g. translations, rotations and scale factors) between each frame of the video sequence, and placing them in a single frame of reference to generate a so-called panoramic image. The operator can then navigate in this panoramic image and find his way more easily, for example, to regions of interest common to another panoramic image produced earlier. This article describes the main image registration methods used in endoscopy, which form the basis of mosaicking. Endoscopic images are often low-contrast and textured, requiring specific and robust registration algorithms.

Finally, the mosaic itself is presented, in particular the placement of non-redundant information in a common reference...