The Performance of Small Support Spatial and Temporal Filters for Dim Point Target Detection in Infrared Image Sequences

The effectiveness of small support spatial filters based on mean, median and morphological opening for the detection of scintillating ultra-dim stationary point targets in IR image sequences has been investigated. The effectiveness of two temporal filters was also studied. The filters were applied to two IR image sequences; an aircraft approaching in an uncluttered background, and an aircraft receding in a bright cloudy background. The spatial filters were effective in detecting the target in the benign background, but neither the spatial nor one of the temporal filters were effective in the cluttered environment. A combination of absolute frame differencing and small support spatial filtering to correct for sensor motion was found to give sufficient increase in signal to clutter ratio to allow detection. RELEASE LIMITATION Approved for public release DEPARTMENT OF DEFENCE DEFENCE SCIENCE & TECHNOLOGY ORGANISATION DSTO Publislied by DSTO Aeronautical and Maritime Research Laboratory 506 Lorimer St Fislwrmans Bend, Victoria 3207 Australia Telephone: (03) 9626 7000 Fax: (03) 9626 7999 © Commonwealth of Australia 2002 AR-012-165 February 2002 APPROVED FOR PUBLIC RELEASE The Performance of Small Support Spatial and Temporal Filters for Dim Point Target Detection in IR Image Sequences Executive Summary Infrared search and track (IRST) systems are being developed to detect anti-ship seaskimming missiles at maximum possible range. The problem is essentially the detection of ultra-dim, stationary, point-like targets in clutter in sequences of infrared (IR) images. This paper considers the worst case when the background consists of low sunlit clouds where the brightness range is very large and cloud edges would provide many artefacts which could be considered as possible targets. The detection of ultra-dim targets poses greater difficulties than the detection of brighter targets. A common method of target detection based on small size spatial filters works by predicting the background by some type of smoothing operation, then subtracting the predicted background from the original image. The aim is to fit the predicted background as closely as possible in the original image without diminishing the target signal. Linear adaptive matched filters can also enhance the response to the target. There is a problem with all spatial filters for detection of ultra-dim point targets with single pixel features in strong clutter because there is no spatial information in a single pixel to allow discrimination from clutter. If the target is moving, filters based on the temporal profile of the target as it moves through a pixel can be used. The target for IRST has been assumed to be stationary, so these temporal filters cannot be used. However, the average brightness of the target is expected to increase over time as it approaches the sensor. In the short time frame there will be considerable variation in brightness due to scintillation. These temporal behaviours may be able to be exploited for target detection. The threat posed by anti-ship missiles is of very short duration, and requires response in shortest possible time. Future staring IRST sensor frame rates are expected to be of the order of 25-50 frames/s, and so the images must be processed in very short times. Hence is essential that detection algorithms have low computational load. The operation of the sensor at high frame rates allows the fact that the target is scintillating to be exploited to improve detectability. Absolute frame differencing would suppress the relatively constant background and highlight the rapid fluctuations in the target brightness cause by scintillation. However, sensor jitter would also produce clutter in the high brightness regions in the image. A combination of absolute frame differencing and small support spatial filtering to correct for sensor motion was found to give sufficient increase in signal to clutter ratio to allow timely detection.