セルフキャリブレーションとNN学習によるカラーテクスチャ物体のレンダリング(一般セッション,光と色の解析・表現とその応用)

概要

論文の詳細を見る
This paper describes a new approach for self-calibration and color image rendering using radial basis function (RBF) neural network. Most empirical approaches make use of a calibration object. Here, we require no calibration object to both shape recovery and color image rendering. The neural network learning data are obtained through the rotations of a target object. The approach can generate realistic virtual images without any calibration object which has the same reflectance properties as the target object. The proposed approach uses a neural network to obtain both surface orientation and albedo, and applies another neural network to generate virtual images from any viewpoint under any direction of light source. Experiments with real data are demonstrated.
2009-10-15