Graph Transform Optimization with Application to Image Compression

Giulia Fracastoro,Dorina Thanou,Pascal Frossard

DOI: https://doi.org/10.48550/arXiv.1712.06393

2019-07-30

Abstract:In this paper, we propose a new graph-based transform and illustrate its potential application to signal compression. Our approach relies on the careful design of a graph that optimizes the overall rate-distortion performance through an effective graph-based transform. We introduce a novel graph estimation algorithm, which uncovers the connectivities between the graph signal values by taking into consideration the coding of both the signal and the graph topology in rate-distortion terms. In particular, we introduce a novel coding solution for the graph by treating the edge weights as another graph signal that lies on the dual graph. Then, the cost of the graph description is introduced in the optimization problem by minimizing the sparsity of the coefficients of its graph Fourier transform (GFT) on the dual graph. In this way, we obtain a convex optimization problem whose solution defines an efficient transform coding strategy. The proposed technique is a general framework that can be applied to different types of signals, and we show two possible application fields, namely natural image coding and piecewise smooth image coding. The experimental results show that the proposed graph-based transform outperforms classical fixed transforms such as DCT for both natural and piecewise smooth images. In the case of depth map coding, the obtained results are even comparable to the state-of-the-art graph-based coding method, that are specifically designed for depth map images.

Information Theory,Image and Video Processing

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to design a new graph transformation method to optimize the rate - distortion performance in signal compression. Specifically, the authors propose a novel graph estimation algorithm to reveal the connectivity between graph signal values by considering the effects of signal and graph topology coding in terms of rate - distortion. In addition, they introduce an innovative graph coding method, that is, regarding the edge weights as another graph signal, which is located on the dual graph. By minimizing the sparsity of the graph Fourier transform (GFT) coefficients of these edge - weight signals on the dual graph, the cost of graph description is introduced into the optimization problem. Eventually, this method can provide an effective transform coding strategy, which is suitable for the compression of different types of signals, especially the compression of natural images and piecewise - smooth images.

### Main Problems
1. **Graph Design and Transformation**: How to design a graph and its corresponding transformation in order to obtain effective rate - distortion performance in signal compression.
2. **Graph Coding Cost**: How to effectively code the graph itself so that it occupies the least number of bits during transmission.
3. **Signal Compression**: How to use the graph transformation method to improve the compression efficiency of signals (especially image signals), especially in the compression of natural images and piecewise - smooth images.

### Solutions
1. **Graph Estimation Algorithm**: A new graph estimation algorithm is proposed to optimize the graph design by considering the rate - distortion performance of signal and graph topology coding.
2. **Graph Coding Method**: Regard the edge weights of the graph as another graph signal, which is located on the dual graph, and control the cost of graph coding by minimizing the sparsity of the GFT coefficients of these edge - weight signals on the dual graph.
3. **Optimization Problem**: Model the graph selection problem as a rate - distortion optimization problem, and find the optimal graph by minimizing the smoothness of signal transformation coefficients and the cost of graph description.

### Formulas
- **Graph Laplacian Matrix**:
\[
L = D - W
\]
where \(D\) is a diagonal matrix, and its \(i\)-th diagonal element \(D_{ii}\) is the sum of all edge weights of node \(i\).

- **Graph Fourier Transform**:
\[
\hat{x}=\Psi^{T}x
\]
where \(\Psi\) is the eigenvector matrix of the graph Laplacian matrix \(L\).

- **Rate - Distortion Optimization Problem**:
\[
\min_{w\in\mathbb{R}^{M}}u^{T}B\mathrm{diag}(w)B^{T}u+\alpha\|\Psi_{d}^{T}w\|_{1}-\beta1^{T}\log(w)
\]
where \(\alpha\) and \(\beta\) are regularization parameters, and \(1\) is an all - one vector.

### Applications
- **Natural Image Compression**: Experimental results show that the proposed graph transformation method is superior to the classical fixed - transform (such as DCT) in natural image compression.
- **Piecewise - Smooth Image Compression**: In piecewise - smooth image compression, this method significantly improves the performance of the classical DCT, and even achieves results comparable to those of the specially - designed graph - based coding method in depth - map coding.

### Conclusion
This paper proposes a new graph - transformation - optimization method, which realizes efficient signal compression by comprehensively considering the coding costs of signals and graph topologies. This method is not only applicable to natural images and piecewise - smooth images, but can also be extended to other types of signal - compression applications.