Lesley Northam

Consistent stylization and painterly rendering of stereoscopic 3D images

We present a method for stylizing stereoscopic 3D images that guarantees consistency between the left and right views. Our method decomposes the left and right views of an input image into discretized disparity layers and merges the corresponding layers from the left and right views into a single layer where stylization takes place. We then construct new stylized left and right views by compositing portions of the stylized layers. Because the left and right views come from the same source layers, our method eliminates common artifacts that cause viewer discomfort. We also present a stereoscopic 3D painterly rendering algorithm tailored to our layer-based approach. This method uses disparity information to assist in stroke creation so that strokes follow surface geometry without ignoring painted surface patterns. Finally, we conduct a user study that demonstrates that our approach to stereoscopic 3D image stylization leads to images that are more comfortable to view than those created using other techniques.

Stereoscopic 3D image stylization

We present a method for stylizing stereoscopic 3D images that guarantees consistency between the left and right views. Our method decomposes the left and right views of an input image into discretized disparity layers and merges the corresponding layers from the left and right views into a single layer where stylization takes place. We then construct new stylized left and right views by compositing portions of the stylized layers. Because the new left and right views come from the same stylized source layers, our method eliminates common stylization artifacts that cause viewer discomfort. We also present a stereoscopic 3D painterly rendering algorithm tailored to our layer-based approach. This method uses disparity information to assist in stroke creation so that strokes follow surface geometry without ignoring painted surface patterns. Finally, we conduct a user study that demonstrates that our approach to stereoscopic 3D image stylization leads to images that are more comfortable to view than those created using other techniques. Graphical abstractWe present a method for stylizing stereoscopic 3D images that guarantees consistency between left and right views.Display Omitted Highlights? A method for stylizing stereo 3D images that guarantees consistency between views. ? Our method works with many stylization filters, including Adobe Photoshop filters. ? We present a stereo 3D painterly rendering algorithm tailored to our approach. ? We present user study results that demonstrate that users prefer our method to others.

A novel first principles approach for the estimation of the sieve factor of blood samples.

Light may traverse a turbid material, such as blood, without encountering any of its pigment containing structures, a phenomenon known as sieve effect. This phenomenon may result in a decrease in the amount of light absorbed by the material. Accordingly, the corresponding sieve factor needs to be accounted for in optical investigations aimed at the derivation of blood biophysical properties from light transmittance measurements. The existing procedures used for its estimation either lack the flexibility required for practical applications or are based on general formulas that incorporate other light and matter interaction phenomena such as detour (scattering) effects. In this paper, a ray optics framework is proposed to estimate the sieve factor for blood samples. It employs a first principles approach to account for the distribution, orientation and shape of the cells that contain hemoglobin, the essential (oxygen-carrying) pigment found in human blood. Within this framework, ray-casting techniques are used to determine the probability that light can traverse a blood sample without encountering any of these cells. The predictive capabilities of the proposed framework are demonstrated through a series of in silico experiments. Its effectiveness is further illustrated by visualizations depicting the different blood parameterizations considered in the simulations.

A collaborative real time previsualization tool for video games and film

The video game and film industries use previsualization (previs) to produce preliminary animations, walk-throughs and still images of scenes, levels and other assets. Using these collaborative planning tools early in the production pipeline reduces costs later on by providing a consistent and detailed cinematographic vision. Most of the previs literature addresses the problem of camera calibration and shot planning [Nitsche 2008]. Gouchet et al. [2007] developed a natural interface for filming a virtual scene using a real camera. Mori et al. [2011] presented an on-set stereoscopic previs method that employs 3D models of characters and sets to calibrate and plan 3D shots.

Ray tracing in the cloud using MapReduce

We present the Hadoop Online Ray Tracer (HORT), a scalable ray tracing framework for general, pay-as-you-go, cloud computing services. Using MapReduce, HORT partitions the computational workload and scene data differently than other distributed memory ray tracing frameworks. We show that this unique partitioning significantly bounds the data replication costs and inter-process communication. Consequently HORT is fault-tolerant and cost-effective when rendering large-scale scenes (i.e., scenes that do not fit into local memory) without specific or dedicated high performance infrastructure. Our experiments demonstrate this scalability and fault tolerance using several CPU and GPU instances on Amazon AWS with the Hadoop open-source implementation of MapReduce.

Brush stroke ordering techniques for painterly rendering

Painterly rendering algorithms often mimic classical hand-painting techniques to automatically generate stylized paintings from input images. These algorithms use a combination of techniques to express a variety of styles and artistic properties (e.g., contrast, mood), but often restrict the user from controlling the rendering order of overlapping brush strokes. This paper illustrates the importance of brush stroke ordering in creating stylistic effects and presents a layer-based painterly rendering algorithm that allows the user to specify a brush stroke ordering. Several of the presented orderings enable the renderer to reduce detail obstruction, simulate handpainting techniques and enhance artistic styles.

HORT: Hadoop online ray tracing with mapreduce

High quality computer-generated imagery (CGI) rendering is a demanding computational task [Hearn and Baker 1997]. To generate high-quality CGI, film studios rely on expensive, specialized, rendering clusters (render farms) [Rath 2009]. Render farms cost a premium over more general infrastructure such as infrastructure as a service (IaaS) offerings. Furthermore, render farms often provide limited selection of commercial rendering applications and have significant overhead with respect to requesting a job. This contrasts general IaaS offerings such as Amazons Elastic Compute Cloud (EC2), which allows for straight-forward, automated provisioning of many virtualized machine instances.

Adjusting the disparity of stereoscopic 3D media in post production

Selecting the convergence point and interaxial distance before the production of stereoscopic 3D media presents several challenges. There are few tools for previsualizing or validating model calculations, so estimating these parameters to match an intended shot relies on costly experience, intuition and iteration. Poor selection of these parameters can also cause or amplify viewer discomfort [Mendiburu 2009]. Figure 1 illustrates the relationship between discomfort and disparity (input images by Scharstein and Pal [2007]).

Consistent stylization of stereoscopic 3D images

The application of stylization filters to photographs is common, Instagram being a popular recent example. These image manipulation applications work great for 2D images. However, stereoscopic 3D cameras are increasingly available to consumers (Nintendo 3DS, Fuji W3 3D, HTC Evo 3D). How will users apply these same stylizations to stereoscopic images?

Artistic sketching with a painterly rendering algorithm

Hand-drawn sketches often depict geometry, colour and texture using loose and roughly drawn lines. Many automated sketching algorithms focus on accurately depicting salient details using pen-and-ink drawings. The approach of Hertzmann et. al. [2000] sketches the contours and silhouettes of 3D meshes, while the interactive algorithm of Kalnins et. al. [2002] renders decorative lines with artistic brushes and suggestions. Other 2D algorithms render Sobel and Canny edges with artistic brushes [Orzan et al. 2007].