PROJECT TITLE :

Depth Super resolution by Transduction - 2015

ABSTRACT:

This paper presents a depth superresolution (SR) method that uses each of a coffee-resolution (LR) depth image and a high-resolution (HR) intensity image. We tend to formulate depth SR as a graph-primarily based transduction problem. In particular, the HR intensity image is represented as an undirected graph, in which pixels are characterized as vertices, and their relations are encoded as an affinity perform. When the vertices initially labeled with sure depth hypotheses (from the LR depth image) are regarded as input queries, all the vertices are scored with respect to the relevances to those queries by a classifying function. Each vertex is then labeled with the depth hypothesis that receives the highest relevance score. We tend to design the classifying operate by considering the local and world structures of the HR intensity image. This approach allows us to handle a depth bleeding drawback that usually appears in current depth SR ways. Furthermore, input queries are assigned in a very probabilistic manner, making depth SR sturdy to noisy depth measurements. We additionally analyze existing depth SR ways within the context of transduction, and discuss their theoretic relations. Intensive experiments demonstrate the prevalence of the proposed technique over state-of-the-art ways both qualitatively and quantitatively.