Regional Foremost Matching for Internet Scene Images

摘要

We analyze the dense matching problem for Internet scene imagesrnbased on the fact that commonly only part of images can bernmatched due to the variation of view angle, motion, objects, etc. Wernthus propose regional foremost matching to reject outlier matchingrnpoints while still producing dense high-quality correspondence inrnthe remaining foremost regions. Our system initializes sparse correspondence,rnpropagates matching with model fitting and optimization,rnand detects foremost regions robustly. We apply our methodrnto several applications, including time-lapse sequence generation,rnInternet photo composition, automatic image morphing, and automaticrnrephotography.