基於梯度域的立體影像仿製

Abstract

本論文提出一個創新的基於梯度域的立體影像仿製技術， 透過同 時完成顏色的調和和形狀的調整， 製作出不管在顏色的表現以及外 觀的形狀上都呈現自然且無縫的合成立體影像。我們的演算法核心 是透過交替進行以下兩個步驟來達到反覆的視差(disparity)調整: 1.基 於梯度域的視差估計。這個步驟為了讓仿製區域的邊界上的視差值 能保持連續， 會每次利用梯度域來重新估計視差值。2.透視感知的 變形法(perspective-aware warping)。這個步驟會根據更新後的視差值重 新調整仿製區域的大小以及形狀。以上的這個程序會能夠確保仿製 區域的邊界上深度的連續，並且塑造出透視法的縮小效果(perspective foreshortening)，使得合成的立體影像更為自然。我們的方法允許不需 要精準切割的物體，如透明，擁有複雜輪廓以及模糊不清的邊界等等 性質的物體被簡單的仿製。在這篇論文的最後我們會利用包括了水面 上的物體，煙火以及彩虹等等具有挑戰性的例子來展示我們的方法。

This thesis presents a novel gradient-domain stereoscopic image cloning technique which performs both color blending and shape adjustment so that the stereoscopic composite is seamless in both color appearance and the perceived shape. The core of the proposed method is an iterative disparity adaptation process which alternates between two steps: gradient-domain disparity estimation, which re-estimates the disparities in the gradient domain so that the disparities are continuous across the boundary of the cloned region; and perspective-aware warping, which re-adjusts the shape and size of the cloned region according to the updated disparities. This procedure guarantees not only depth continuity across the boundary but also models perspective foreshortening, leading to more natural stereoscopic composites. Our method allows cloning objects with transparency, intricate silhouette and vague boundaries easily because it does not require precise segmentation of the objects. We demonstrates stereoscopic image cloning on several challenging cases including water surface, firework and rainbow.