Machine Vision and Applications (2002) 13: 109–110 Machine Vision and Applications c  Springer-Verlag 2002 Editorial Introduction to the Special Issue on Innovative Applications of Computer Vision The computer vision and pattern recognition field has been very active and is witnessing a multitude of techniques be- ing developed for a variety of applications. The challenge has been to successfully develop robust vision systems that can work effectively in real-world situations. The special is- sue presents seven diverse applications of computer vision and pattern recognition. The papers presented in this special issue were selected from the 36 papers that were presented (and selected from 90 submitted papers) in a single track at the 5th IEEE Workshop on Applications of Computer Vision (WACV2000) held in Palm Springs, California, in December 2000. The first paper in this special issue is by Appel and Navab ontheregistrationoftechnicaldrawingsandcalibratedimages for industrial augmented reality. The paper presents an algo- rithm to fully integrate technical drawings, calibrated images andas-built3Dmodels.Thenewalgorithmrequiresonlyafew correspondences between points on a technical drawing and multiple images in order to estimate a metric planar transfor- mation between the drawing and the world coordinate system. It describes the mathematical relationship between this trans- formationandthesetofhomographiesneededformergingthe technicaldrawingwithallthecalibratedimages.Thetechnical drawings are superimposed on the photogrammetric images, thus significantly improving the industrial planning process. The second paper by Joyeux et al. is on the tracking and map reconstruction of line scratches in degraded motion pic- tures. It describes a working system for automatically detect- ing and removing line scratch artifacts from digitized old and grainy black and white films. It has a mix of application- specificrequirementsandapproachesusingtechniquessuchas Kalman filtering over multiple frames and a Bayesian restora- tion technique to achieve the reconstruction of damaged areas in motion picture films. The third paper by Ortiz et al. is on a vision system for trackingunderwatercablesatvideorates.Theoperatingcondi- tions, guiding a stable underwater vehicle, and environmental conditions such as sand, algae, rocks, etc. surrounding the ca- blesmaketheirdetectionandtrackingadifficultproblem.The paper describes a method based on image segmentation, cable detection (position and orientation) and comparison of mea- surementswithprediction.Ifthemeasurementsandprediction for the region-of-interest do not agree, a recovery mechanism isactivated.Resultsareshownonseveralsequencesofvarying quality. The fourth paper by Beveridge et al. presents an approach for automated velocity picking in semblance velocity images (velocity vs. time images) for geophysical data interpretation. Velocity picking is one of the important steps used to develop a map of the earth’s crust from acoustic data collected on the surface. The authors present a two-step process where in the first step the intensity peaks are found in the semblance velocity image and in the second step a local search method, guided by the median of the result of the first step, is used to find a locally optimal subset of peaks that represent the most coherent curve through the image. The authors claim performanceaccuracycomparabletothatofthehumanexpert. The fifth paper by Liu andVemuri is on the efficient mul- timodal image registration. Both the rigid motion and non- rigid deformation cases are considered. The paper presents a combination of techniques for multimodal registration, com- bining a local frequency image representation based on Gabor wavelets with a quasi-Newton and a PDE-based morphing technique that estimates the non-rigid alignment. Examples are presented using computed tomography (CT) and magnetic resonance (MR) scan images. ThenextpaperbyRussakoffandHermanisonheadtrack- ing using stereo. The authors propose an approach for head tracking by using stereo depth images instead of conventional intensity images. First, the stereo depth images are used for humanfiguresegmentation,takingadvantageofthecharacter- istic that the stereo depth data are insensitive to color, shadow, orlightingvariation.Toaddresstheproblemthatthedepthdata are sensitive to image textures, a surface validation procedure is applied in the segmentation algorithm. Then local edge de- tectors are employed on the extracted foreground to find the occludingshoulderedgesthatareusedforfittingasimpletorso model. The location and orientation of the head are easily es- timated by the model parameters. Because the procedure can be performed on each frame separately, it is less sensitive to rapid movements or temporary occlusions than conventional approaches. The experiments show that this model achieves good results in locating the head. The last paper by Bradski and Davis is on motion segmen- tation and pose recognition using a motion history gradient method to represent motion in an image sequence. This paper