Scene Recognition with Infrared, Low-Light, and Sensor-Fused Imagery

摘要

The goal of this study was to evaluate the information conveyed by single- and dual-band sensorimagery by assessing performance on a scene recognition task. An experiment tested immediaterecognition for pictures following an initial brief viewing. Images were taken with an uncooled LIMIRSlong-wave infrared sensor and a Fairchild image intensified low-light CCD, and were compared againstfused, false-color images created by remapping both spectral bands into a two-dimensional color space(Scribner et al., 1996). In each trial, an image was presented for 100 msec and followed immediately bya 300-msec checkerboard mask. A second image, of the same or of a different sensor format, was thendisplayed and remained visible until a response was made. The observer’s task was to indicate whetherthe first and second image depicted the same scene, regardless of which sensor format the scenes weredisplayed in. All possible permutations of sensor formats were used. It was hypothesized that colorfusion,which combined and potentially enhanced information contained in the single-band images,would allow more accurate scene recognition. Performance was best when the first and second imageswere presented in the same format. When format changed between the presentation of the two images,performance deteriorated, but more so when the second image was of a single band format. Format ofthe first image itself had little effect, indicating that the primary benefits of sensor fusion were inmatching the content of the second image to a stored representation of the first, and not in processing thebriefly viewed first image. These results suggest that fusion can allow information from multiple singlebandsensors to be effectively combined and presented within a single image, within which componentinformation remains perceptually accessible.