Michael J. Horowitz

Last updated

Michael J. Horowitz (born January 2, 1964, in Ames, Iowa) is an American electrical engineer whose contributions to video coding standardization and early integration of newly minted standards into video products have facilitated the widespread commercial adoption of the H.264|AVC, H.264 SVC, H.265|HEVC, and AV1 video coding standards. He is co-inventor of flexible macroblock ordering (FMO) [1] and tiles [2] , essential features in H.264|AVC and H.265|HEVC, respectively.

Contents

Early Background

Horowitz was born and raised in Ames, Iowa. He received an A.B. degree from Cornell University. In 1998, under the supervision of Professor David Neuhoff [3] , he received a Ph.D. in electrical engineering from The University of Michigan [4] .

Standardization

Several video coding standards have been published over the past forty years. Not all have been widely adopted. Excellent compression performance does not guarantee commercial viability. Other factors such as computational complexity play important roles. Horowitz led efforts to reduce computational complexity during the development of the H.264|AVC [5] [6] and H.265|HEVC standards [7] . In a related effort, to leverage the full computational power of multi-core CPUs, he led the effort to develop H.265|HEVC coding tools for high-level parallelism [8] (e.g., tiles).  

In addition to computational complexity reduction work, Horowitz has led standardization development efforts for error resilience tools [9] (e.g., FMO in H.264|AVC) and the real-time bitstream transport protocol (RTP) for AV1 [10] .

Early Commercialization

The complexity of modern video coding standards demands a great deal of engineering effort and skill to develop product-ready implementations of the standard. For example, to become production-ready a software codec implementation may need to be optimized to run thousands of times faster than the standard’s reference software. Products released shortly after the publication of a standard demonstrate commercial viability thereby facilitating adoption by other manufacturers who might otherwise opt not to risk using the standard.

In May of 2000 while at Polycom, Horowitz developed and demonstrated the first product-ready implementation of macroblock-adaptive multiple reference frames [11] overcoming concerns that its high computational complexity would prevent commercial adoption. The tool was published as Annex U of H.263. Macroblock-adaptive multiple reference frames is a mainstay of every subsequent video coding standard.

In 2003, while still at Polycom, he led the team that produced the first commercially available in-product implementation of H.264|AVC [12] demonstrating that the new H.264|AVC standard could be used successfully in real-time low-delay commercial products.

In 2008, at Vidyo, Horowitz architected and led the team that developed the first commercially available in-product implementation of H.264 SVC [13]  

In 2012, at eBrisk Video, Horowitz served as lead engineer of the team that developed one of the first commercially available implementations of H.265|HEVC demonstrating that a real-time software implementation of the standard could yield high coding efficiency. [14] .

In 2020, at Google, Horowitz served as technical lead of the AV1 for Google Duo project that became the first commercially available real-time interactive video implementation of AV1, dispelling the myth that older entry-level mobile devices did not have sufficient computational resources to support a real-time software implementation of AV1 [15] . The AV1 implementation in Duo and subsequently Google Meet supports video calling at bit rates as low as 40 kilobits per second on modest entry-level mobile devices thereby enabling people to connect in regions of the world where video calls would not otherwise be possible.

References

  1. United States Patent 7,239,662
  2. United States Patent 9,060,174
  3. "David Neuhoff". web.eecs.umich.edu. Retrieved December 1, 2025.
  4. Horowitz, Michael J.; Neuhoff, David L. (July 17, 1998). "Image coding by perceptual pruning with a cortical snapshot indistinguishability criterion". SPIE Proceedings. 3299. SPIE: 330–339. doi:10.1117/12.320154.
  5. https://www.itu.int/wftp3/av-arch/video-site/0109_San/VCEG-N06.doc [ bare URL DOC/DOCX file ]
  6. https://www.itu.int/wftp3/av-arch/jvt-site/2002_01_Geneva/JVT-B012.doc [ bare URL DOC/DOCX file ]
  7. https://www.itu.int/wftp3/av-arch/video-site/1001_Kyo/VCEG-AM02.zip [ bare URL non-HTML file ]
  8. https://www.itu.int/wftp3/av-arch/jctvc-site/2012_02_H_SanJose/JCTVC-H_Notes_dI.doc [ bare URL DOC/DOCX file ]
  9. https://www.itu.int/wftp3/av-arch/jvt-site/2002_05_Fairfax/JVT-C002d5.doc [ bare URL DOC/DOCX file ]
  10. "RTP Payload Format For AV1". aomediacodec.github.io. Retrieved December 1, 2025.
  11. Wiegand, T., Girod, B. “Multi-frame motion-compensated prediction for video transmission”, page xi, Springer, 2001.
  12. Andrew W. Davis, "The Wainhouse Bulletin," Volume 4 Issue #8, February 2003.
  13. "IP Video Conferencing | Desktop Video Conferencing | Vidyo". Archived from the original on May 7, 2009. Retrieved January 7, 2014.
  14. Horowitz, M., Kossentini, F., Mahdi, N., Xu, S., Guermazi H., Tmar, H., Li B., Sullivan, G. J., Xu, J., "Informal subjective quality comparison of video compression performance of the HEVC and H.264/MPEG-4 AVC standards for low-delay applications”, Proc. SPIE 8499, Applications of Digital Image Processing XXXV, October 15, 2012.
  15. Nick Statt, "Google Duo video calls are about to look a whole lot better," The Verge, April 22, 2020.