Rendering Semitransparent Layered Media

March 14, 2007 at 4:04 pm (Rendering Semi-Transparent Layered Media)

In this project we are going to implement the algorithm presented at “Rendering Semitransparent Layered Media” [Oat2006]. Modern graphics hardware provides the capability to apply many texture layers onto a single surface. However, simply alpha blending many textures onto a surface does not give the appearance of surface depth that one would expect from semitransparent surface layers. Normal mapping, transparency masking, offset mapping and image filtering are combined in this method to produce realistic looking layered surfaces.

The method presented in the paper achieves the look of a volumetric material by exploiting several perceptual cues, based on depth and illumination, while combining multiple material layers on the surface of an otherwise non-volumetric, multi-textured surface such as the human heart shown in the first page. Multiple implementation strategies are suggested that allow for different trade-offs to be made between visual quality and runtime performance and we are going to investigate which one works best for our implementation.

You can download the design doc here

Permalink Leave a Comment

Undergrad Thesis – “Adding Mutative Programming Structures To Java”

February 23, 2007 at 6:38 pm (Projects)

This is the thesis I did as an undergraduate at the University of Piraeus, Greece. It’s in Greek though so I don’t know if it can be of any use to you.

Download it here.

Permalink Leave a Comment

Machine Learning – Algorithm Comparisons

February 23, 2007 at 6:29 pm (Projects)

In this paper I compare the performance of K – Nearest Neighbors, Streamwise Linear Regression (SLR) and a combination of the two, first without and then with PCA. The datasets I used are The Insurance Company Benchmark dataset and the Johns Hopkins University Ionosphere database.

It is interesting to see how kNN performs in comparison to SLR when PCA is not used and the results we get when it is. Especially in the second case, data rescaling is introduced which affects kNN and SLR in a different degree. This is because kNN is not scale invariant, while on the other hand SLR is. The combined method adopts the best of the two methods and provides better results.

Download my paper here.

Permalink Leave a Comment

The Direct3D 10 System

February 23, 2007 at 6:16 pm (Projects)

Download my presentation here.

Permalink Leave a Comment

Fire & Smoke Simulation in Computer Graphics

February 23, 2007 at 1:16 pm (Projects)

Download my presentation here.

Permalink Leave a Comment

« Previous page