Brain-inspired cortical-coding algorithm for multimedia processing

Abstract

This thesis presents an innovative approach to multimedia data compression, drawing inspiration from the human brain's neocortex. The study addresses the need for advanced compression techniques in response to the growing volumes of multimedia data. This study begins with an extensive literature review that sets the context by examining the limitations of existing compression methods, particularly standard lossy codecs. It also explores the emerging potential of neural codecs, establishing a theoretical basis for the development of a new, brain-inspired compression algorithm. This algorithm aims to surpass current methods in compression efficiency, quality of decompression, and processing speed. In the methodology section, the thesis describes the design and implementation of the novel cortical-coding algorithm, which mimics the neocortex's method of processing information. The experimental framework is carefully detailed, including the theoretical underpinnings and specific algorithms employed to benchmark the codec's performance against both traditional and neural codecs. The results obtained are promising, showing that the cortical coding algorithm competes with and excels beyond selected traditional codecs (MP3, AAC, OPUS, OGG Vorbis) and neural codecs (EnCodec, SoundStream) in several key performance metrics. These findings are analyzed in depth, highlighting significant advancements in compression ratio, and output quality, while showing real-time processing capability. The discussion delves into the broader implications of these results, particularly their potential impact on real-time multimedia applications such as video conferencing, live streaming, and virtual reality. It is posited that the successful application of biomimetic principles with the proposed codec design can revolutionize multimedia data handling, providing more efficient and scalable solutions. The thesis is concluded by summarizing the research contributions, which include the successful demonstration of a novel, efficient, and effective approach to data compression, mainly audio and image compression, inspired by cortical coding principles. Recommendations for future research include further refinement of the codec and exploration into its application across different multimedia types to enhance versatility and utility. This thesis provides important new insights into multimedia compression and suggests new possibilities for applying neuroscience in developing digital technologies. It sets the stage for further interdisciplinary research that has the potential to impact the field of multimedia data processing significantly.