Abstract. Gradient-domain compositing is an essential tool in computer vision and its applications, e.g., seamless cloning, panorama stitching, shadow removal, scene completion and reshuffling. While easy to implement, these gradient-domain techniques often generate bleeding artifacts where the composited image regions do not match. One option is to modify the region boundary to minimize such mismatches. However, this option may not always be sufficient or applicable, e.g., the user or algorithm may not allow the selection to be altered. We propose a new approach to gradient-domain compositing that is robust to inaccuracies and prevents color bleeding without changing the boundary location. Our approach improves standard gradient-domain compositing in two ways. First, we define the boundary gradients such that the produced gradient field is nearly integrable. Second, we control the integration process to concentrate residuals where they are less conspicuous. We show that our approach can be formulated as a standard least-squares problem that can be solved with a sparse linear system akin to the classical Poisson equation. We demonstrate results on a variety of scenes. The visual quality and run-time complexity compares favorably to other approaches. Key words: gradient-domain compositing, visual masking 1

Figure 1: Left: Shadow volumes are extruded in a geometry shader and granular visibility queries are used directly on the GPU to exclude unlit objects from the shadow volume generation. Right: Visibility determination on the GPU is also beneficial in costly rendering techniques such as displacement mapping to cull individual, occluded primitives. Efficient visibility queries are key in many interactive rendering techniques, such as occlusion culling, level of detail determination, and perceptual rendering. The occlusion query mechanism natively supported by GPUs is carried out for batches of rendered geometry. In this paper, we present two novel ways of determining visibility by intelligently querying summed area tables and computing a variant of item buffers. This enables visibility queries of finer granularity, e.g., for sub-regions of objects and for instances created within a single draw call. Our method determines the visibility of a large number of objects simultaneously which can be used in geometry shaders to cull triangles, or to control the level of detail in geometry and pixel shaders under certain rendering scenarios. We demonstrate the benefits of our method with two different real-time rendering techniques.

We propose a simple algorithm to control the spatial location of reconstruction errors inherent in gradient-based image compositing. We build upon the classical Poisson equation and add a weighting term that controls where reconstruction errors can occur. We define this term in such a way that residuals are mainly located in textured regions where they are less visible. Our approach is independent of how the composited gradient field has been built and is complementary to the methods that focus on this aspect. Our approach retains the simplicity of the traditional Poisson equation while producing more pleasing composites. 1.

Real-time realistic rendering of a computer generated scene is one of the core research areas in computer graphics as it is required in several applications such as computer games, training simulators, medical and architectural packages and many other fields. The key factor of realism in the rendered images is the simulation of light transport based on the given lighting conditions. More natural results are achieved using luminance values near to the physical ones. However, the vast range of real luminances has a far greater range of values than what can be displayed on standard monitors. As a final step to the rendering process, a tonemapping operator needs to be applied in order to transform the values in the rendered image to displayable ones. Illumination of a scene is usually approximated with the rendering equation which solution is a computational expensive process. Moreover, the computational cost increases even more with the increase in the number of light sources and the number of vertices of the objects in the scene. Furthermore, in order to achieve high frame rates, current illumination algorithms compromise the quality with assumptions for several factors or assume static scenes so that they can exploit precomputations.

We present a quasi-continuous level of detail method that is based on an image error metric to minimize the visual error. The method is designed for objects of high geometric complexity such as trees. By successive simplifications, it constructs a level of detail hierarchy of unconnected primitives (ellipsoids, lines) to approximate the input models at increasingly coarser levels. The hierarchy is constructed automatically without manual intervention. When rendering roughly 100k model instances at a low visual error compared to rendering the full resolution model, our method is two times faster than billboard clouds. Categories and Subject Descriptors (according to ACM CCS): I.3.3 [Computer Graphics]: Picture/Image Generation- Display algorithms I.3.6 [Computer Graphics]: Methodology and Techniques- Graphics data structures and data types 1.

We present an optimized pruning algorithm that allows for considerable geometry reduction in large botanical scenes while maintaining high and coherent rendering quality. We improve upon previous techniques by applying model-specific geometry reduction functions and optimized scaling functions. For this we introduce the use of Precision and Recall (PR) as a measure of quality to rendering and show how PR-scores can be used to predict better scaling values. We conducted a user-study letting subjects adjust the scaling value, which shows that the predicted scaling matches the preferred ones. Finally, we extend the originally purely stochastic geometry prioritization for pruning to account for view-optimized geometry selection, which allows to take global scene information, such as occlusion, into consideration. We demonstrate our method for the rendering of scenes with thousands of complex tree models in real-time.