固有空間への非線形射影に基づく画像各部の注目度の定量化

概要

論文の詳細を見る
This paper describes a method that estimates the degree of attractivity of each region of an input image. The method consists of two stages; one is the learning stage and the other is the estimation stage. In the learning stage, comparatively low dimensional eigen space is determined by Principal Component Analysis (PCA) of many samples of learning images which are taken by a fixed camera. In the estimation stage, on the other hand, the input image is projected on the eigen space, and the residue is estimated as an image that represents the degree of attractivity of each region of the input image. In the proposed method the projection onto the eigen space is calculated non-linearly so as to obtain a more appropriate image for representing the degree of attractivity, the effect of which is verified by experiments.
社団法人映像情報メディア学会の論文
2005-06-01