Comparative Assessment of ICA Architectures for Gait Recognition

摘要

ICA expands on traditional data transformation and dimensionality reduction techniques and can be used for better efficiency and enhanced discrimination when comparing with other techniques, e.g., PCA and LDA. We compare the performance on normalized gait sequences using both local ICA (architecture I) and global ICA (architecture II). This is also evaluated against competing techniques such as PCA and LDA. FastICA on ICA architecture II produces results that are competitive with state-of-the art methods on gait recognition from different viewpoints (experiment A from the gait recognition challenge.)