Wide-Angle Rectification via Content-Aware Conformal Mapping

Input wide-angle image (e) Image content analysis (h) Our rectification (c) Perspective projection (g) Shih's rectification (b) Mercator projection (f) Carroll's rectification (d) Stereographic projection Figure 1. Given a wide-angle image (as shown in (a) e.g.), this paper proposes a new image rectification method which automatically corrects evident wide-angle lens distortions (e.g. curved ceiling lines, and skewed face), and obtains a rectified image (e.g. (h)) . This is achieved by deep-learning based image content analysis ((e)), as well as content-aware conformal mapping. Our method effectively reduces wide-angle distortions, and at the same time maintains the original ultra-wide field of view, without sacrificing salient image contents. For comparison, we also show results by previous methods in (b,c,d,f,g). Our method achieves the best results. For more comparisons please see the paper as well as the supplemental material. Better viewed on screen with zooming in.