View Interpolation Improvement Using Suitable Constraints of Feature Tracking Parameters

概要

論文の詳細を見る
View interpolation has been extensively studied in recent years. It is considered as synthesizing a series of continuous and realistic intermediate views between a pair of reference images without 3D knowledge. In general, the synthesizing procedure consists of five main steps including feature tracking, fundamental matrix constructing, image pre-warping, view morphing, and image post-warping. Because the last three steps are formulated as a common approach, the first two steps determine the accuracy and efficiency of view interpolation. In this paper, we discuss the reasons and phenomena by which the feature tracking parameters and their change constraints can influence the fundamental matrix constructing. Here, the discussed feature tracking parameters are mainly set by focal length, feature number, and feature patch size. Enlarging focal length, increasing feature number, and setting suitable feature patch size can generate the higher quality correspondences of feature, and then compute the correct fundamental matrix. These parameters' influences in the first step could be transferred to the other steps of view interpolation, then determine the final appearance of view interpolation. Experimental results show that in acceptable computational complexity, the feature tracking parameters can change the visual effects in view interpolation.
一般社団法人画像電子学会の論文