无线多媒体传感器网络中基于低秩和深度神经网络的高效、非对称、鲁棒图像编码传输

In the wireless multimedia sensor network(WMSN) where the front-end sensor nodes have limited computing power and the channel is unstable, how to transmit images efficiently and robustly is a huge challenge. Considering the traditional joint source and channel coding is inefficient in the above mentioned applications, we propose an asymmetric (the encoder side is simple and the decoder side can be relatively complicated), robust and efficient image coding transmission scheme based on compressive sensing(CS).. Particularly, at the encoder side, the compression algorithm based on random interleaving and blocking is used to avoid the complex entropy coding and channel coding in the traditional joint source and channel coding, and the simplicity and robustness of the encoder are realized. A non-uniform progressive quantizer based on the look-up table is designed to achieve higher rate-distortion performance. At the decoder side, the similarity among the image blocks is used, and the non-local low rank approximation algorithm is used to reconstruct an initial images. Since the complex texture is often lost in the low rank reconstruction, a generative adversarial networks(GAN) in Shearlet domain is proposed to restore the lost texture. . This project will provide an important theoretical reference for image coding transmission in an asymmetric and unsteady network environment.

在前端传感器节点计算能力有限，信道传输网络不可靠的无线多媒体传感器网络中，如何高效鲁棒地传输图像面临巨大的挑战。考虑到传统联合信源信道编码在上述场合效率低下，本申请课题拟在压缩感知理论框架下，基于非局部低秩逼近和深度神经网络模型，提出具有非对称（编码器端简单、解码器端可相对复杂）、鲁棒、高效这三大特点的图像编码传输方案。. 具体来说，在编码端，采用随机交织分块的压缩算法，避免了复杂的熵编码和信道编码环节，以实现编码器的简单性和鲁棒性；为了提升率失真性能，针对随机测量值之间的相关性和统计分布特性，设计一种基于查找表的渐进式非均匀量化器；在解码端，先采用非局部低秩策略重建出原始图像的低秩逼近样本；考虑到低秩逼近中容易丢失复杂纹理信息，再借助于Sheatlet域的生成对抗网络模型来重构丢失的纹理信息。. 本申请项目将为非对称、非稳定网络环境下的图像编码传输提供重要参考。

在前端传感器节点计算能力有限，传输信道不可靠的无线多媒体传感器网络中，如何高效鲁棒地传输图像面临巨大的挑战。考虑到传统联合信源信道编码在上述场合效率低，本项目旨在压缩感知理论框架下，研究具有非对称（解码器端可相对复杂但编码器要简单）、鲁棒和高效这三大特点的图像编码传输方案。. 本项目对基于压缩感知的图像编码方法进行了全面研究，分析了现有方法的优缺点，开展了高性能压缩感知随机测量算法设计、高效率量化算法设计，以及结合图像局部和非局部先验的基于数据驱动的高性能重建算法设计，据此构建了一种具有非对称、鲁棒和高效特性的适合于无线传感器网络的图像压缩编码传输方案，实验验证了所提算法的有效性，所提算法在编码简单性、鲁棒性和高效性上大幅度超越现有JPEG2000+信道编码的传统图像编码传输方案。本项目从理论上探索了基于压缩感知的图像编码算法性能，所提算法还可以推广到无人机、深空探测图传中，具有广阔的应用前景。