Real-time TV commercial monitoring based on robust visual hashing

This paper presents a method for TV commercial monitoring that utilize matching visual extracts to detect known commercials. These visual extracts, also known as visual hashes, are computed from the low-frequency part of both luminance and chroma video frame components. Our approach relies on the assumption that commercial breaks are delimited by known commercial bumpers. Video sequences containing unknown commercials are easily detected and recognized later by manual inspection. Our method was implemented in a prototype application and achieved very high detection rate while false acceptance and false rejection rates were quite low.