Robust visual correspondence: theory and applications

Visual correspondence represents one of the most important tasks in computer vision. Given two sets of pixels (i.e. two images), it aims at finding corresponding pixel pairs belonging to the two sets (homologous pixels). As a matter of fact, visual correspondence is commonly employed in fields such as stereo correspondence, change detection, image registration, motion estimation, pattern matching, image vector quantization. The visual correspondence task can be extremely challenging in presence of disturbance factors which typically affect images. A common source of disturbances can be related to photometric distortions between the images under comparison. These can be ascribed to the camera sensors employed in the image acquisition process (due to dynamic variations of camera parameters such as auto-exposure and auto-gain, or to the use of different cameras), or can be induced by external factors such as changes of the amount of light emitted by the sources or viewing of non-lambertian surfaces at different angles. All of these factors tend to produce brightness changes in corresponding pixels of the two images that can not be neglected in real applications implying visual correspondence between images acquired from different spatial points (e.g. stereo vision) and/or different time instants (e.g. pattern matching, change detection). In addition to photometric distortions, differences between corresponding pixels can also be due to the noise introduced by camera sensors. Finally, the acquisition of images from different spatial points or different time instants can also induce occlusions. Evaluation assessments have also been proposed which compared visual correspondence approaches for tasks such as stereo correspondence (Chambon & Crouzil, 2003), image registration (Zitova & Flusser, 2003) and image motion (Giachetti, 2000)