Metadata in Sequential Real-Time 2-D Detection

Extraction of patterns is an important low-level operation in several vision applications. In particular, this work is motivated by the problem of detecting unknown vague structures in images with a minimum sample number. Tracing of an unknown structure is allowed by the spatial 2-D placement of extremum. Often, this permits near-optimum testing of hypothesis, which also accentuates the descriptive value of extrema. The proposed sequential approach guarantees predictable detection delay, which is a significant advantage in real-time use. The performance is demonstrated with simulations and real experiments, where transients have unknown starting time