Universitätsbibliothek
Weimar Open Access

Replacing a uniform illumination by a high-frequent illumination enhances the contrast of observed and captured images. We modulate spatially and temporally multiplexed (projected) light with reflective or transmissive matter to achieve high dynamic range visualizations of radiological images on printed paper or ePaper, and to boost the optical contrast of images viewed or imaged with light microscopes.

The advent of high-performance mobile phones has opened up the opportunity to develop new context-aware applications for everyday life. In particular, applications for context-aware information retrieval in conjunction with image-based object recognition have become a focal area of recent research. In this thesis we introduce an adaptive mobile museum guidance system that allows visitors in a museum to identify exhibits by taking a picture with their mobile phone. Besides approaches to object recognition, we present different adaptation techniques that improve classification performance. After providing a comprehensive background of context-aware mobile information systems in general, we present an on-device object recognition algorithm and show how its classification performance can be improved by capturing multiple images of a single exhibit. To accomplish this, we combine the classification results of the individual pictures and consider the perspective relations among the retrieved database images. In order to identify multiple exhibits in pictures we present an approach that uses the spatial relationships among the objects in images. They make it possible to infer and validate the locations of undetected objects relative to the detected ones and additionally improve classification performance. To cope with environmental influences, we introduce an adaptation technique that establishes ad-hoc wireless networks among the visitors’ mobile devices to exchange classification data. This ensures constant classification rates under varying illumination levels and changing object placement. Finally, in addition to localization using RF-technology, we present an adaptation technique that uses user-generated spatio-temporal pathway data for person movement prediction. Based on the history of previously visited exhibits, the algorithm determines possible future locations and incorporates these predictions into the object classification process. This increases classification performance and offers benefits comparable to traditional localization approaches but without the need for additional hardware. Through multiple field studies and laboratory experiments we demonstrate the benefits of each approach and show how they influence the overall classification rate.

CAMShift is a well-established and fundamental algorithm for kernel-based visual object tracking. While it performs well with objects that have a simple and constant appearance, it is not robust in more complex cases. As it solely relies on back projected probabilities it can fail in cases when the object's appearance changes (e.g. due to object or camera movement, or due to lighting changes), when similarly colored objects have to be re-detected or when they cross their trajectories. We propose extensions to CAMShift that address and resolve all of these problems. They allow the accumulation of multiple histograms to model more complex object appearance and the continuous monitoring of object identi- ties to handle ambiguous cases of partial or full occlusion. Most steps of our method are carried out on the GPU for achieving real-time tracking of multiple targets simultaneously. We explain an ecient GPU implementations of histogram generation, probability back projection, im- age moments computations, and histogram intersection. All of these techniques make full use of a GPU's high parallelization.

We present a novel image classification technique for detecting multiple objects (called subobjects) in a single image. In addition to image classifiers, we apply spatial relationships among the subobjects to verify and to predict locations of detected and undetected subobjects, respectively. By continuously refining the spatial relationships throughout the detection process, even locations of completely occluded exhibits can be determined. Finally, all detected subobjects are labeled and the user can select the object of interest for retrieving corresponding multimedia information. This approach is applied in the context of PhoneGuide, an adaptive museum guidance system for camera-equipped mobile phones. We show that the recognition of subobjects using spatial relationships is up to 68% faster than related approaches without spatial relationships. Results of a field experiment in a local museum illustrate that unexperienced users reach an average recognition rate for subobjects of 85.6% under realistic conditions.

We present a simple and cost-efficient way of extending contrast, perceived tonal resolution, and the color space of static hardcopy images, beyond the capabilities of hardcopy devices or low-dynamic range displays alone. A calibrated projector-camera system is applied for automatic registration, scanning and superimposition of hardcopies. We explain how high-dynamic range content can be split for linear devices with different capabilities, how luminance quantization can be optimized with respect to the non-linear response of the human visual system as well as for the discrete nature of the applied modulation devices; and how inverse tone-mapping can be adapted in case only untreated hardcopies and softcopies (such as regular photographs) are available. We believe that our approach has the potential to complement hardcopy-based technologies, such as X-ray prints for filmless imaging, in domains that operate with high quality static image content, like radiology and other medical fields, or astronomy.