A baseline object detection algorithm using background anomalies for electro-optic identification sensors

Electro-optic identification sensors provide photographic quantity that can be used to identify mine-like contacts provided by long-range sensors, such as sonar systems. Computer aided identification and automatic target recognition algorithms have been funded by the Office of Naval Research (322-OP) to help operators in the identification process. Under this funding, a new background anomaly approach was introduced (Nevis et al., 2002) for object detection using a least square error (LSE) background best fit searching for anomalies of certain specified size and shape (and SNR) that stick out from the local background. This initial approach has been refined into a baseline algorithm. An overview of this baseline algorithm is presented in this paper, with an emphasis on modifications from the initial approach.