ICAフィルタとガボール関数の符号化の比較

概要

論文の詳細を見る
The independent component analysis of natural scenes spans filters similar to real-valued Gabor functions. However, ICA filters also present fluctuations that diverge from Gabor functions. In this research, we investigate the role of these components in the efficient coding of natural scenes. Here, we assume a model where ICA filters are represented as the convolution of a Gabor component with a function of unknown characteristics. Then, we analyze how the coding behavior of ICA filters differs from that of their Gabor components. Specifically, we examine characteristics of the coding such as representation sparseness and error. This initial research suggests that using only Gabor functions without fluctuations and irregularities alters the image representation, which may have implications for general sensory stimuli coding.
2011-02-28