Applying manifold learning techniques to the CAESAR database

A limited comparative study of dimension reduction techniques on CAESAR

“…An OpenGL viewer was created to view the LiDAR scan models in the native .ply format. Results of the data using this viewer are shown in Figure 2 from [3]. Included in the dataset are demographic data for each individual including characteristics such as gender, ethnicity, and age, detailed reports describing the collection and preprocessing of the data, and source code for some functions to manipulate the data.…”

“…To reduce discrepancies between the North American and European databases as described in [19], only the North American database was used in our work and, as described in [3], a total of 22 subjects were removed due to missing data. The resulting database was comprised of 2369 subjects with 1250 females (52.76% of the database) and 1119 males (47.24% of the database).…”

“…What is the purpose of reducing the dimensionality of a dataset? As stated in [3]: efficient processing, visualization, and data collection reduction. Efficient processing, as described in [4], refers to combating the 'curse of dimensionality'.…”

“…Furthermore, there are other dimension reduction techniques that need to be considered. Thus, the goal of this effort is to expand the experiments in [3]. We will use two dimension reduction techniques, Diffusion Map and Principal Component Analysis (PCA), and two classifiers, KNN and Adaboost, and compare the results.…”

“…In [3] a dimension reduction technique was applied to the CAESAR database which showed that it is possible to reduce the original dimensionality to three with reasonably successful classification accuracy. However, in that work, they only used one dimension reduction technique and a single classifier: Diffusion Map and KNN classifier.…”