Extension Of Guided Filter With Memory For Projection On Training Spectrum

Myounghwan Kim, Jungi Lee

Abstract: Hyperspectral imaging which has a wide range of bands has been employed in various industries. Hyperspectral imaging has recently been employed not only for classification but also for anomaly detection. In anomaly detection, several false alarms are occurred due to the noise data. In this paper, we proposed memory guided filter that consists of guided filter and memory that contains sparse normal. The memory stores representative spectra from training data, and then the guided filter projects data to a similar spectrum in memory. Our proposed method reduces noise by projection and outperforms the conventional methods, demonstrating the robustness of memory guided filter.