In this paper, we propose a novel multi-view method for surface reconstruction from matched line segments with applications to robotic mapping and image-based rendering. Starting from 3D line segments and poses obtained via Line-SLAM, we project segments from multiple frames into keyframes for image-space analysis. For each keyframe, a grid of image faces is created by optimized intersection of the segment projection lines. These faces define a segment adjacency graph, wherein we perform our Breadth-First Plane Search (BFPS). The found plane hypotheses are merged maximally with respect to a structure-preserving criterion by growing coplanar regions across the graph. Hypotheses violating the visibility constraint are discarded based on fast per-face and mostly nongeometrical evaluation of the scene and image graph. Finally, each image face gets back-projected onto an optimal plane to obtain a 3D surface model. The presented system is a complete and automatic solution suitable for mapping an environment in real-time scenarios like robotic exploration. We demonstrate the performance of our algorithm on several indoor scenes of varying complexity. Compared to a pure 3D analysis of segments, we see a speed-up by one to almost two orders of magnitude, while still improving on reconstruction accuracy.

The complexity in face recognition emerges from the variability of the appearance of a human face. While the identity is preserved, the appearance of a face may change due to factors such as illumination, pose or facial expression. To recognize a person independent of pose, we want to separate shape from texture information. We concentrate on the texture part in this work. We first fit an active appearance model to a given facial image. The shape information is used to transform the face into a shape-free representation. We decompose the transformed face into local regions and extract texture features from these not necessarily rectangular regions using a shape-adapted discrete cosine transform. The texture features we use for face recognition are independent of pose and shape of the face. We show that these features contain sufficient discriminative information to recognize persons across changes in pose.

Lip motion analysis can enhance a security system based on face recognition significantly. Examining speaker dependent lip movements and visually determining the spoken content can prevent attacks using photographs or pre-recorded video sequences. Our system operates under active near infrared illumination. We investigate the use of a new type of features, namely local binary patterns, to model lip motions with hidden Markov models. We evaluate the classification accuracy with the TUNIR database, which we made available to the public for the future comparison of competing approaches.

In this paper we present an approach for lip motion analysis that can be used in conjunction with a person authentication system based on face recognition, to avoid attacks on the system using passive photographs. This work focuses on robustly tracking lips in gray scale images, which may be captured in the visible light or near infrared spectrum. We present an approach for locating the two lip corners in a face image. Then we extract suitable features from the mouth region to classify mouth states (visemes). The system shows a classification accuracy of above 85%. The temporal changes in the detected viseme classes can be used for detecting the imposter.

We present a model for the lips using the active shape model (ASM) based on Bezier curves which can capture the dynamics of the lips efficiently. The curves are defined only by a few points that can be used as visual speech features. Using some measured feature points, we can recover the full model of the lips especially for those features which are difficult to be detected. Accurate detection of features of the lips is implemented using multiple independent feature templates.

Stereoscopic calibration and reconstruction is applied to the specialized optics of a binocular monobjective stereo light microscope. Such a microscope exhibits a special kind of image distortion. Despite the difficulty of modelling the microscope, a simple calibration method as well as a fast and simple, yet precise, reconstruction algorithm is developed. Their fundamental scheme is based upon biological binocular vision. The reconstruction uses polynomial approximations up to a degree of 2 and thus has a very low computational complexity. The polynomial coefficients are identified during calibration and their number is minimal by construction. No lens data is required. Both the calibration and reconstruction algorithm are robust against a rigid motion of the microscope. Their power is proven with real data using an off-the-shelf PC.

In this paper, we present a technique for the extraction of the five main visemes produced in natural speech for German. The method belongs to the LDA (Linear Discriminant Analysis) family. The intensity, the edges, and the line segments are used to locate the lips automatically and for viseme classification. Using many features in the recognition maximizes the probability of recognition rate. The corners of the mouth are used in case of small rotation and scale. An experiment has been carried out on different people, to understand the part of the speech that the human being use. The people grouped the phonemes into five different visemes. The number of distinguished visemes is not the same for each speaker. Everyone express the speech in a different visemes. Good recognition rate has been achieved on different speaker.

We present a high quality solution to the binarization problem of technical document images. The method is based on locally adaptive methods without the need of manual interaction or tuning. The algorithm includes Niblack's (1986) binarization and two validation steps based on morphological image processing and gradient based decisions. Our result has the following advantages. Alphanumeric labels are restored, even if they are hardly readable in the source image. Parallel, multiple lines with small distance are reconstructed. Even detail of very low contrast in the vicinity of strong contrast image areas is reconstructed (e.g. faint stamp lines). Variable background intensity is suppressed, but texture and large font bold face letters are restored.

This paper presents an algorithm for segmentation of image sequences where specially aspects for object oriented coding are taken into account. A fundamental requirement of such applications is the temporal stability of the segmentation. This is improved compared to other existing approaches by including motion estimation into the segmentation process. Additionally, a hierarchical approach enables an efficient predictive coding on one hand and a semantic data access on the other hand. As a direct result from using full colour information for the segmentation process, coding of the chrominance information can be done with extremely high compression ratios. Thus there is nearly no extra coding effort for colour images compared to greyscale images.