View Dependent Stochastic Sampling for Efficient Rendering of Point Sampled Surfaces

Sushil Bhakar, Liang Luo, S. P. Mudur

Concordia University

Computer Science Department

Montreal H3G1M8

Canada

email: {sushi_bh | liang_lu | mudur}@cs.concordia.ca http://www.cs.concordia.ca/~faculty/mudur

Abstract

In this paper we present a new technique for rendering very large datasets representing point-sampled surfaces. Rendering efficiency is considerably improved by using stochastic sampling that is controlled using various object and view dependent image space properties. Most of the current rendering algorithms simplify the model in a preprocessing step before rendering. This simplification primarily results in a smaller subset of sampled points. Hence these algorithms suffer from the problem of under-sampling when the screen space resolution becomes greater than the sampling rate inherent in the simplified representation. Our algorithm avoids this problem by accessing the original point data set at all times and dynamically selecting points to display at rendering time. As a side benefit our preprocessing is much simpler and preprocessing time is also considerably reduced, albeit at the cost of increased disk and memory usage. We also include an algorithm to correctly estimate properly oriented normals, which are essential during rendering.