Rate Efficient Visual Correspondences
Kannan Ramchandran, Chuohao Yeo and Parvez Ahammad
National Science Foundation and Agency for Science, Technology and Research (Singapore)
Establishing visual correspondences is a critical step in a myriad of computer vision tasks, especially those involving multiple views of a scene, as is the case in wireless camera networks. In a constantly changing environment and when cameras are mobile, visual correspondences need to be updated on a recurring basis. Furthermore, the use of wireless links between camera motes impose tight rate constraints. In order to avoid a single point of failure and to reduce the redundant computational costs involved in computing robust feature descriptors for all received images at every node, distributed approaches are appealing in the context of camera networks. Traditional centralized approaches are neither suitable in a distributed setting nor take into account overall communication costs. These issues motivate us to consider the problem of establishing visual correspondences in a distributed fashion between cameras operating under rate constraints via communicating compact descriptors.
First, we verify that descriptors of regions which are in correspondence are highly correlated and propose a novel use of distributed source coding to reduce the overall bandwidth necessary for exchanging feature descriptors required to establish correspondences across camera views. Second, we propose and describe a systematic approach for evaluating relative performance of our proposed and baseline schemes by studying their rate vs. visual correspondence retrieval performance.
Our evaluations demonstrate that the proposed scheme is able to provide compression gains of up to 76% with minimal loss in the number of correctly established correspondences compared to a scheme that communicates the entire image of the scene losslessly in compressed form. The proposed distributed source coding framework also provides superior performance when compared to simply transmitting the feature descriptors independently without exploiting the availability of correlated descriptors of corresponding features.
Figure 1: Problem setup
- C. Yeo, P. Ahammad, and K. Ramchandran, "A Rate-Efficient Approach for Establishing Visual Correspondences via Distributed Source Coding," SPIE Visual Communications and Image Processing, 2008.
- C. Yeo, P. Ahammad, and K. Ramchandran, "Rate-Efficient Visual Correspondences Using Random Projections," IEEE International Conference in Image Processing, 2008.