Information visualization

In this dissertation we apply sparse constraints to improve optical flow and
trajectories. We apply sparsity in two ways. First, with 2-frame optical flow, we
enforce a sparse representation of flow patches using a learned overcomplete dictionary. Second, we apply a low rank constraint to trajectories via robust coupling. We begin with a review of optical flow fundamentals. We discuss the commonly used flow estimation strategies and the advantages and shortcomings of each. We introduce the concepts associated with sparsity including dictionaries and low rank matrices.

Molecular dynamics is a computer simulation technique for expressing the
ultimate details of individual particle motions and can be used in many fields, such as
chemical physics, materials science, and the modeling of biomolecules. In this thesis, we
study visualization and pattern mining in molecular dynamics simulation. The molecular
data set has a large number of atoms in each frame and range of frames. The features of
the data set include atom ID; frame number; position in x, y, and z plane; charge; and
mass. The three main challenges of this thesis are to display a larger number of atoms and
range of frames, to visualize this large data set in 3-dimension, and to cluster the
abnormally shifting atoms that move with the same pace and direction in different frames.
Focusing on these three challenges, there are three contributions of this thesis. First, we
design an abnormal pattern mining and visualization framework for molecular dynamics
simulation. The proposed framework can visualize the clusters of abnormal shifting atom
groups in a three-dimensional space, and show their temporal relationships. Second, we propose a pattern mining method to detect abnormal atom groups which share similar
movement and have large variance compared to the majority atoms. We propose a
general molecular dynamics simulation tool, which can visualize a large number of atoms,
including their movement and temporal relationships, to help domain experts study
molecular dynamics simulation results. The main functions for this visualization and
pattern mining tool include atom number, cluster visualization, search across different
frames, multiple frame range search, frame range switch, and line demonstration for atom
motions in different frames. Therefore, this visualization and pattern mining tool can be
used in the field of chemical physics, materials science, and the modeling of
biomolecules for the molecular dynamic simulation outcomes.

Number perception, its neural basis and its relationship to how numerical stimuli are presented have been challenging research topics in cognitive neuroscience for many years. A primary question that has been addressed is whether the perception of the quantity of a visually presented number stimulus is dissociable from its early visual perception. The present study examined the possible influence of visual quality judgment on quantity judgments of numbers. To address this issue, volunteer adult subjects performed a mental number comparison task in which two-digit stimulus numbers (Arabic number format), among the numbers between 31 and 99 were mentally compared to a memorized reference number, 65. Reaction times (RTs) and neurophysiological (i.e. electroencephalographic (EEG) data) responses were acquired simultaneously during performance of the two-digit number comparison task. In this particular quantity comparison task, the number stimuli were classified into three distance factors. That is, numbers were a close, medium or far distance from the reference number (i.e., 65). In order to evaluate the relationship between numerical stimulus quantity and quality, the number stimuli were embedded in varying degrees of a typical visual noise form, known as "salt and pepper noise" (e.g., the visual noise one perceives when viewing a photograph taken with a dusty camera lens). In this manner, the visual noise permitted visual quality to be manipulated across three levels: no noise, medium noise (approximately 60% degraded visual quality from nonoise), and dense noise (75% degraded visual quality from no-noise).

Natural matte extraction is a difficult and generally unsolved problem. Generating a matte from a nonuniform background traditionally requires a tediously hand drawn matte. This thesis studies recent methods requiring the user to place only modest scribbles identifying the foreground and the background. This research demonstrates a new GPU-based implementation of the recently introduced Fuzzy- Matte algorithm. Interactive matte extraction was achieved on a CUDA enabled G80 graphics processor. Experimental results demonstrate improved performance over the previous CPU based version. In depth analysis of experimental data from the GPU and the CPU implementations are provided. The design challenges of porting a variant of Dijkstra's shortest distance algorithm to a parallel processor are considered.