Effective gradient domain object editing on mobile devices

  • Authors:
  • Yingen Xiong;Dingding Liu;Kari Pulli

  • Affiliations:
  • Nokia Research Center, Palo Alto, CA;Nokia Research Center, Palo Alto, CA and University of Washington, Seattle, WA;Nokia Research Center, Palo Alto, CA

  • Venue:
  • Asilomar'09 Proceedings of the 43rd Asilomar conference on Signals, systems and computers
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

We present a gradient domain object editing approach and its implementation for mobile devices. It can be used for creating a new composite image by removing, adding, and moving objects in an image. The approach can be divided into two parts: creation and editing of a new gradient vector field, and recovery of a new composite image from the new gradient vector field. In the first part, a new gradient vector field is created from the gradients of the source image, and then updated by inserting the new object gradients, by removing object gradients and filling removed areas with the gradients of best-fit patches found in other parts of the source image, or by combining these two processes when moving objects. In the second part, a divergence vector field is computed from the gradient vector field and used for a guidance vector to construct a Poisson equation. The new composite image is recovered from the gradient vector field by solving the Poisson equation with boundary conditions. Our approach can merge all regions in the picture seamlessly with smooth color transition for the whole picture. It can be used for large object removal and for filling the background of the removed object. The final composite image is a globally optimal solution. The approach is implemented and runs with good performance on mobile camera phones.