Analyzing Near-infrared Images for Utility Assessment

Visual cognition is of significant importance in certain imaging applications, such as security and surveillance. In these applications, an important issue is to determine the cognition threshold, which is the maximum distortion level that can be applied to the images while still ensuring that enough information is conveyed to recognize the scene. The cognition task is usually studied with images that represent the scene in the visible part of the spectrum. In this paper, our goal is to evaluate the usefulness of another scene representation. To this end, we study the performance of near-infrared (NIR) images in cognition. Since surface reflections in the NIR part of the spectrum is material dependent, an object made of a specific material is more probable to have uniform response in the NIR images. Consequently, edges in the NIR images are likely to correspond to the physical boundaries of the objects, which are considered to be the most useful information for cognition. This feature of the NIR images leads to the hypothesis that NIR is better than a visible scene representation to be used in cognition tasks. To test this hypothesis, we compared the cognition thresholds of NIR and visible images performing a subjective study on 11 scenes. The images were compressed with different compression factors using JPEG2000 compression. The results of this subjective test show that recognizing 8 out of the 11 scenes is significantly easier based on the NIR images when compared to their visible counterparts.