We present a first approach to light-field retargeting using z-stack seam carving, which allows light-field compression and extension while retaining angular consistency. Our algorithm first converts an input light field into a set of perspective-sheared focal stacks. It then applies 3D deconvolution to convert the focal stacks into z-stacks, and seam-carves the z-stack of the center perspective. The computed seams of the center perspective are sheared and applied to the z-stacks of all off-center perspectives. Finally, the carved z-stacks are converted back into the perspective images of the output light field. To our knowledge, this is the first approach to light-field retargeting. Unlike existing stereo-pair retargeting or 3D retargeting techniques, it does not require depth information.

With the continuously increasing sensor resolutions of cameras, light field imaging is becoming a more and more practical extension to conventional digital photography. It complements postprocessing by synthetic aperture control, refocusing, as well as perspective and field-of-view changes. For being a true alternative to classical 2D imaging, however, the spatial resolution of light fields must be in the same megapixel-order as the resolution of today’s digital images. The additional angular resolution must also be adequately high to prevent sampling artifacts (in particular for synthetic re-focussing). This will quickly cause gigabytes rather than megabytes of data that have to be rendered with limited graphics memory. We describe a light-field caching framework that makes it possible to render very large light fields in real-time.

With increasing sensor resolutions of digital cameras, light-field imaging is becoming more and more relevant, and might even replace classical 2D imaging in photography sooner or later. It enables, for instance, digital refocussing and perspective changes after capturing. Rescaling light fields to different resolutions and aspect rations, however, is challenging. As for regular image and video content, a linear scaling alters the aspect ratio of recorded objects in an unnatural way. In contrast, image and video retargeting utilizes a nonlinear and content-based scaling. Applying image retargeting to individual video frames independently does not retain temporal consistency. Similarly, applying image retargeting naively to the spatial domain of light fields will not retain angular consistency. We present a first approach to light-field retargeting. It allows compressing or stretching light-fields while retaining angular consistency.

We show how the intrinsically performed JPEG compression of many digital still cameras leaves margin for deriving and applying image-adapted coded apertures that support retention of the most important frequencies after compression. These coded apertures, together with subsequently applied image processing, enable a higher light throughput than corresponding circular apertures, while preserving adjusted focus, depth of field, and bokeh. Higher light throughput leads to proportionally higher signal-to-noise ratios and reduced compression noise, or -alternatively- to lower shutter times. We explain how adaptive coded apertures can be computed quickly, how they can be applied in lenses by using binary spatial light modulators, and how a resulting coded bokeh can be transformed into a common radial one.

A variety of standard video modes that stretch or zoom lower resolution video content linearly to take full advantage of large screen sizes have been implemented in TV sets. When content and screen aspect ratios differ, format proportions may be compromised, video content may be clipped, or screen regions may remain unused. Newer techniques, such as video retargeting and video upsampling, rescale individual video frames and can potentially match them to the display resolution and aspect ratio. However, none of these methods can display simultaneously more than is contained in a single frame.

We present a new video mode for television sets that we refer to as display pixel caching (DPC). It fills empty borders with spatially and temporally consistent information while preserving the original video format. Unlike related video modes, such as stretching, zooming and video retargeting, DPC does not scale or stretch individual frames. Instead, it merges the motion information from many subsequent frames to generate screen filling panoramas in a consistent manner. In contrast to state-of-the-art video mosaicing, DPC achieves real-time rates for high resolution video content while processing more complex motion patterns fully automatically. We compare DPC to related video modes in the context of a user evaluation.

With the increasing computational capacity of camera-equipped mobile phones, object recognition on such devices is shifting away from centralized client-server approaches, in which the phones act only as input/output front-ends, to local on-device classification systems. The advantages of such a decentralization are shorter response times, scalability with respect to a large number of simultaneous users, and reduced network traffic costs. Mobile image classification can support applications that rely on device localization, such as museum or city guidance, by supplementing existing positional information retrieved, for instance, from GPS or
GSM cells. The challenge for mobile image classification, however, is to become as robust as possible, even when applied in large, highly dynamic, and uncontrollable public environments: Hundreds to thousands of objects must be recognized from different perspectives, from varying distances, and under changing lighting conditions, while recognition rates must remain usable. The key to solving this problem may be automatic adaptation to dynamic changes in the environment and to the most common user behavior.
This paper summarizes the various components of our mobile museum guidance system PhoneGuide.

The combination of advanced software algorithms and optics opens up new possibilities for display, imaging, and lighting. It makes possible responsive optical systems that adapt to particular situations automatically and dynamically. Visual computing is a relatively young research field that provides a foundation for many of these approaches. It represents a tight coupling between image synthesis, image analysis, and visual perception. While optics is all about image formation, visual computing deals with the general processing of images. This paper summarizes several examples that illustrate how graphics, vision, perception, and optics are combined to realize smart projectors, smart cameras, and smart light sources.

We present a multi-image classification technique for mobile phones that is supported by relational reasoning. Users capture a sequence of images employing a simple near-far camera movement. After classifying distinct keyframes using a nearest-neighbor approach the corresponding database images are only considered for a majority voting if they exhibit similar near-far inter-image relations to the captured keyframes. In the context of PhoneGuide, our adaptive mobile museum guidance system, a user study revealed that our multi-image classification technique leads to significantly higher classification rates than single image classification. Furthermore, when using near-far image relations, less keyframes are sufficient for classification. This increases the overall classification speed of our approach by up to 35%.

We present an unobtrusive technique for supporting and improving object recognition approaches on mobile phones. To accomplish this we determine the present and future locations of museum visitors by evaluating user-generated spatio-temporal pathway data. In the context of our adaptive mobile museum guidance system called PhoneGuide we show that this improves the classification performance significantly and can achieve recognition rates comparable to those of traditional location-based image classification approaches. Over a period of four months, we collected the pathway data of 132 regular museum visitors at the Natural History Museum of Erfurt, Germany.

We show that optical inverse tone-mapping (OITM) in light microscopy can improve the visibility of specimens, both when observed directly through the oculars and when imaged with a camera. In contrast to previous microscopy techniques, we pre-modulate the illumination based on the local modulation properties of the specimen itself. We explain how the modulation of uniform white light by a specimen can be estimated in real-time, even though the specimen is continuously but not uniformly illuminated. This information is processed and back-projected constantly, allowing the illumination to be adjusted on the fly if the specimen is moved or the focus or magnification of the microscope is changed. The contrast of the specimen's optical image can be enhanced, and high-intensity highlights can be suppressed. A formal pilot study with users indicates that this optimizes the visibility of spatial structures when observed through the oculars. We also demonstrate that the signal-to-noise (S/N) ratio in digital images of the specimen is higher if captured under an optimized rather than a uniform illumination. In contrast to advanced scanning techniques that maximize the S/N ratio using multiple measurements, our approach is fast because it requires only two images. This can be beneficial for image analysis in digital microscopy applications with real-time capturing demands.

Coding a projector's aperture plane with adaptive patterns together with inverse filtering allow the depth-of-field of projected imagery to be increased. We present two prototypes and corresponding algorithms for static and programmable apertures. We also explain how these patterns can be computed at interactive rates, by taking into account the image content and limitations of the human visual system. Applications such as projector defocus compensation, high quality projector de-pixelation, and increased temporal contrast of projected video sequences can be supported. Coded apertures are a step towards next-generation auto-iris projector lenses.

We present a multi-image classification technique for mobile phones that is supported by relational reasoning. Users capture a sequence of images employing a simple near-far camera movement. After classifying distinct keyframes using a nearest-neighbor approach the corresponding database images are only considered for a majority voting if they exhibit similar near-far inter-image relations to the captured keyframes. In the context of PhoneGuide, our adaptive mobile museum guidance system, a user study revealed that our multi-image classification technique leads to significantly higher classification rates than single image classification. Furthermore, when using near-far image relations, less keyframes are sufficient for classification. This increases the overall classification speed of our approach by up to 35%.

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 low-cost extensions to CAMShift that address and resolve all of these problems. They allow the accumulation of multiple histograms to model more complex object appearances and the continuous monitoring of object identities 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 efficient GPU implementations of histogram generation, probability back projection, computation of image moments, and histogram intersection. All of these techniques make full use of a GPU’s high parallelization capabilities.

We show how temporal backdrops that alternately change their color rapidly at recording rate can aid chroma keying by transforming color spill into a neutral background illumination. Since the chosen colors sum up to white, the chromatic (color) spill component is neutralized when integrating over both backdrop states. Being able to separate both states, however, additionally allows to compute high quality alpha mattes. Besides neutralizing color spill, our method is invariant to foreground colors and supports applications with real-time demands. In this article, we explain different realizations of temporal backdrops and describe how keying and color spill neutralization are carried out, how artifacts resulting from rapid motion can be reduced, and how our approach can be implemented to be compatible with common real-time post-production pipelines.