Digital Video

Visual compression and inference to enable immersive and interactive multimedia.

Digital video research at MERL spans a wide range of topics including compression, image processing and analysis. Our goal is to develop technologies that enable communication of rich and high-quality multimedia data, as well as immersive and interactive multimedia applications.

As video resolution continues to increase, so does the need for compression. We have been actively exploring techniques to improve the coding efficiency of high-resolution video formats, including novel transforms and texture coding methods; perceptual video coding is another active area of study. Our group has also been a leader in the standardization of 3D and multiview video coding formats, including the development of the MVC format that has been adopted for Blu-ray and broadcast applications, as well as next-generation depth-based coding techniques to enable virtual view generation. We have also been involved in research related to various display processing functions including video noise reduction and format conversions.

To satisfy applications that require lightweight encoders and high compression efficiency, such as space and airborne systems, the group has also developed novel coding architectures and techniques based on distributed source coding principles. With this coding approach, complexity is shifted from the encoder to the decoder. Our current research focuses on the coding of multi-spectral image data.

There are also a broad range of applications that require visual features to be communicated and processed with low bit rate and latency, e.g., augmented reality. Our recent work on visual inference makes use of quantized embeddings to determine a compact representation of visual features to perform classification and recognition tasks under these constraints. We are also investigating novel feature spaces that facilitate a broad range of inference tasks.