F-number control for Accurate Depth Estimation with Color-Filtered Aperture

スポンサーリンク

概要

• 論文の詳細を見る

• Accurate depth map estimation is one of the hot topics in the field of computer vision as well as computer graphics, because of its wide range of applications. The color-filtered aperture (CFA) approach has shown to be promising with its handy hardware installation and simple estimation algorithm. However, it does not always perform well for various scene situations, if all the optical parameters are fixed. In this paper, we show that controlling f-number based on the geometrical optics effectively cope with variety of scene situations. More specifically, we first propose a way to estimate the optimal f-number from layout of the camera and objects, e.g., distance between the camera and objects, and relative distances between objects. We also propose a depth estimation from multiple images with different f-numbers, which widens the applicable range of scene variation with high accuracy. We demonstrate the performance gain of our approaches. Our approach can be realized without additional hardware modification if we install the CFA on a lens equipped with an adjustable pupil aperture.

• 2010-11-11

著者

• Hiroki Yazawa

  Optical Research Laboratory, Nikon Corporation

• Shinichi Nakajima

  Optical Research Laboratory, Nikon Corporation

• Hiroki Yazawa

  Optical Research Laboratory Nikon Corporation

• Shinichi Nakajima

  Optical Research Laboratory Nikon Corporation

関連論文

• F-number control for Accurate Depth Estimation with Color-Filtered Aperture
• Non-sparse Feature Mixing in Object Classification
• Exhaustive Search of Feature Subsets for Support Vector Machine Classification

スポンサーリンク