PROJECT TITLE :

Automatic Depth Extraction from 2D Images Using a Cluster-Based Learning Framework - 2018

ABSTRACT:

There has been a important increase in the provision of 3D players and displays in the last years. Nonetheless, the quantity of 3D content has not experimented an increment of such magnitude. To alleviate this problem, many algorithms for changing pictures and videos from 2D to 3D are proposed. Here, we present an automatic learning-based mostly 2D-3D image conversion approach, based on the key hypothesis that color images with similar structure seemingly gift an identical depth structure. The presented algorithm estimates the depth of a color query image using the prior information provided by a repository of color + depth images. The algorithm clusters this database aiming to their structural similarity, and then creates a representative of each color-depth image cluster that can be used as previous depth map. The choice of the acceptable prior depth map corresponding to one given color query image is accomplished by comparing the structural similarity in the color domain between the question image and also the database. The comparison is based on a K-Nearest Neighbor framework that uses a learning procedure to build an adaptive combination of image feature descriptors. The most effective correspondences confirm the cluster, and in turn the associated previous depth map. Finally, this previous estimation is enhanced through a segmentation-guided filtering that obtains the final depth map estimation. This approach has been tested using 2 publicly offered databases, and compared with many state-of-the-art algorithms in order to prove its efficiency.