Godwin Shen

Edge-adaptive transforms for efficient depth map coding

In this work a new set of edge-adaptive transforms (EATs) is presented as an alternative to the standard DCTs used in image and video coding applications. These transforms avoid filtering across edges in each image block, thus, they avoid creating large high frequency coefficients. These transforms are then combined with the DCT in H.264/AVC and a transform mode selection algorithm is used to choose between DCT and EAT in an RD-optimized manner. These transforms are applied to coding depth maps used for view synthesis in a multi-view video coding system, and provides up to 29% bit rate reduction for a fixed quality in the synthesized views.

Edge-aware intra prediction for depth-map coding

This work proposes a new intra prediction coding scheme for depth map images used in view interpolation. The main goal is to design a prediction scheme which can reduce the prediction error energy in blocks with arbitrary edge shapes. This will reduce the rate needed to encode such blocks while also eliminating some of the annoying artifacts caused by quantization. Since depth maps typically consist of smooth regions separated by edges, we find it sufficient to design prediction schemes which can make effective use of edge information. Working from the intra prediction framework in H.264, we provide a graph representation of pixels in a block and pixels from previously coded blocks and construct an edge-aware prediction scheme based on this. We also employ existing rate-distortion (RD) optimization methods to further improve the coding performance. Our proposed methods reduce the bit rate for depth maps by up to 29% for a fixed interpolated PSNR for some sequences.

Optimized distributed 2D transforms for irregularly sampled sensor network grids using wavelet lifting

We address the design and optimization of an energy-efficient lifting-based 2D transform for wireless sensor networks with irregular spatial sampling. The 2D transform is designed to allow for unidirectional computation found in existing path-wise transforms, thereby eliminating costly backward transmissions often required by existing 2D transforms, while simultaneously achieving greater data decorrelation than those path-wise transforms. We also propose a framework for optimizing the 2D transform via an extension of standard dynamic programming (DP) algorithms, where a selection is made among alternative coding schemes (e.g., different number of levels in the wavelet decomposition). A recursive DP formulation is provided and an algorithm is given that finds the minimum cost coding scheme assignment for our proposed 2D transform.

Edge-preserving depth-map coding using graph-based wavelets

In order to efficiently encode depth map images in a multi-view video coding scenario, two basic properties of these images can be leveraged: first, errors in pixels located near the edges of objects have a greater perceptual impact on the synthesized view; second, depth maps can be approximated as piecewise planar signals. We make use of these facts to define a discrete wavelet transform using lifting that avoids filtering across edges. The filters are designed to fit the planar shape of the signal. This leads to an efficient representation of the image while preserving the edge information. By preserving the edge information, we are able to improve the quality of the synthesized views, as compared to existing methods.

Transform-Based Distributed Data Gathering

A general class of unidirectional transforms is presented that can be computed in a distributed manner along an arbitrary routing tree. Additionally, we provide a set of conditions under which these transforms are invertible. These transforms can be computed as data is routed towards the collection (or sink) node in the tree and exploit data correlation between nodes in the tree. Moreover, when used in wireless sensor networks, these transforms can also leverage data received at nodes via broadcast wireless communications. Various constructions of unidirectional transforms are also provided for use in data gathering in wireless sensor networks. New wavelet transforms are also proposed which provide significant improvements over existing unidirectional transforms.

Unidirectional graph-based wavelet transforms for efficient data gathering in sensor networks

We design lifting-based wavelet transforms for any arbitrary communication graph in a wireless sensor network (WSN). Since transmitting raw data bits along the routing trees in WSN usually requires more bits than transmitting encoded data, we seek to minimize raw data transmissions in the network. We especially focus on unidirectional transforms which are computed as data is forwarded towards the sink on a routing tree. We formalize the problem of minimizing the number of raw data transmitting nodes as a weighted set cover problem and provide greedy approximations. We compare our method with existing distributed wavelet transforms on communication graphs. The results validate that our proposed transforms reduce the total energy consumption in the network with respect to existing designs.

Signal compression in wireless sensor networks

Signal compression is an important tool for reducing communication costs and increasing the lifetime of wireless sensor network deployments. In this paper, we overview and classify an array of proposed compression methods, with an emphasis on illustrating the differences between the various approaches.

Adaptive distributed transforms for irregularly sampled Wireless Sensor Networks

We develop energy-efficient, adaptive distributed transforms for data gathering in wireless sensor networks. In particular, we consider a class of unidirectional transforms that are computed as data is forwarded to the sink along a given routing tree and develop a tree-based Karhunen-Loàve Transform (KLT) that is optimal in that it achieves maximum data de-correlation among this class of transforms. As an alternative to this KLT (which incurs communication overhead in order to learn second order data statistics), we propose a backward adaptive filter optimization algorithm for distributed wavelet transforms that i) achieves near optimal performance and ii) has no communication overhead in learning statistics.

Joint Routing and 2D Transform Optimization for Irregular Sensor Network Grids Using Wavelet Lifting

We address the joint optimization of routing and compression for wireless sensor networks using a lifting-based 2D transform that can be computed along arbitrary routing trees. The proposed 2D transform allows for unidirectional computation, thereby eliminating costly backward transmissions often required by existing 2D transforms. We also propose a framework for optimizing the transform by selecting among a different set of coding schemes (i.e., different levels in the wavelet decomposition). Since our transform can operate on arbitrary routing trees, we focus on the problem of jointly optimizing routing trees based on inter-node data correlation and inter-node distance. The two extreme solutions would be i) to route data along paths that maximize inter-node data correlation (at the risk of increasing transport costs), corresponding to a minimum spanning tree (MST), or ii) to follow shortest path tree (SPT) routing (where inter-node data correlation may not be as high). We propose an optimization technique that exhaustively searches for the optimal tree over a set of combinations of MST and SPT. We also propose a heuristic approximation algorithm that is amenable for use on larger networks and with which we observe total cost reductions close to 10% for some of the data.

Tree-based wavelets for image coding: Orthogonalization and tree selection

In this work we consider the design of the lifting filters and trees used in a separable tree-based wavelet transform. We first consider the use of improved prediction filters, optimized to represent more efficiently smooth signals for arbitrary tree structures. We then consider the design of update filters that are orthogonal to neighboring prediction operators. While the corresponding decomposition is not fully orthogonal, near orthogonality between prediction and update operators leads to significant improvements in energy compaction. Finally we consider the design of trees that (i) avoid filtering across discontinuities in an image to reduce the amount of high frequency energy, while (ii) maintaining some regularity in the downsampled grids over multiple levels of decomposition in order to achieve good spatial localization of filtering.