Compressive sensing for image and video processing, coding and communication

当前的图像和视频通信系统以高分辨率获取图像或视频，并且随后有效地压缩信号，丢弃大量数据。这种方法使图像传感器和压缩算法复杂化，并且需要大的读出带宽，特别是在高帧速率下。这里提出的研究重点是一种新的方法，其中压缩在信号采集过程中进行;一种称为压缩传感（CS）的范例。在研究过程中，我们将设计和制作基于CS的图像传感器，并研究CS在图像和视频处理和通信中的应用。有许多应用程序，CS可以是有帮助的。标准视频捕获系统需要一组完整的样本，在该点上可以应用压缩算法。在某些应用中，获得这些原始样品可能是困难、昂贵或耗时的。在其他应用中，例如多相机网络，实现压缩算法本身可能是一个挑战。我们认为，这些负担可以通过使用压缩成像，其中随机测量独立收集，不需要额外的压缩来减少。作为交换，实现这种方案的挑战在于开发有效的稀疏诱导表示和相应的算法，用于从随机测量中恢复视频。该研究将通过开发基于CS的分布式和多视图视频编码、基于CS的检测/分类以及基于CS的图像传感器的设计和制造的新方法来改善图像和视频采集和通信。CS有望大幅提高数据采集、处理和融合系统的性能和能力，同时降低部署的成本和复杂性。有迷人的理论和实践研究问题，在改进的图像和视频采集，处理和通信的有前途的回报，在拟议的研究计划。

Current image and video communications systems acquire the image or video in high resolution and subsequently compress the signal effectively discarding a large amount of data. This approach complicates the image sensor and the compression algorithm and requires large read-out bandwidths especially at high frame rates. The research proposed here focuses on a new approach where compression is carried out during signal acquisition; a paradigm known as compressive sensing (CS). During the course of the research we will design and fabricate CS-based image sensors and study the applications of CS to image and video processing and communication. There are numerous applications where CS could be helpful. Standard video capture systems require a complete set of samples at which point a compression algorithm may be applied. In some applications, it may be difficult, expensive or time consuming to obtain these raw samples. In other applications, such as multi-camera networks, implementing a compression algorithm may itself be a challenge. We argue that these burdens can be reduced by using compressive imaging where random measurements are collected independently and no additional compression is needed. In exchange, the challenge of implementing such a scheme comes in developing efficient sparsity-inducing representations and the corresponding algorithms for video recovery from random measurements. The proposed research will improve image and video acquisition and communication by developing new methods for CS-based distributed and multi-view video coding, CS-based detection/classification and design and fabrication of CS-based image sensors. CS promises to substantially increase the performance and capabilities of data acquisition, processing and fusion systems while lowering the cost and complexity of deployment. There are fascinating theoretical and practical research problems in the proposed research program with promising payoffs in improved image and video acquisition, processing and communications.