Interactive 2D 3D image conversion method for mobile devices

Abstract

This work proposes an interactive tool for creating stereo image from a mono image. The user interaction is defined as scribbling on the object of interest followed by relative depth assignment to the selected object. Initial step in the algorithm is to create structured image oversegments with intensity homogeneity and geometrical convexity constraint. The final image segmentation is realized by merging the oversegmented regions in a region growing manner. The proposed local neighbourhood similarity based method saves the energy of searching globally optimum cut. Instead of utilizing an iterative method to search for an optimum energy minimization, local neighbourhood based energy calculation enables one step decision. As the proposed method has relatively low complexity, it could efficiently be used in mobile devices with limited computational resources. The assigned relative depths of multiple objects are used to create stereo image pairs using conventional depth image based rendering (DIBR). The tool has been implemented for Nokia N900 Phone and allows creating stereo images with positive and negative disparity.