Image feature utilization for target detection in monochromatic images

This dissertation is a study of the key image qualities that people can use to perform complex image discrimination tasks, and discusses how these findings might be applied. Three specific scenarios are examined.; The first is an investigation into the specific cues for looming object detection. Textured surfaces elicit a greater sense of "motion in depth" (or looming) and horizontal expansion of an object is the optimal direction for accurate detection. This information can be used to develop a driver-assistance warning signal that may reduce rear-end collisions.; The second scenario is an examination of the strategies observers use to detect grating patterns in noise through a technique called classification images. The classification images revealed non ideal behavior with no summation for the combined 1 and 3cpd stimuli. This has interesting implications for the various models for multiple channel grating detection.; The third scenario is a study to quantify the image features that inspectors could be using to detect threat items in x-ray images of luggage. A list of image features was generated and tested for utility. The features that generated the most optimal performance were then used as part of a proposed new training procedure. The results showed that the training could produce a performance level that is above chance, and only take a fraction of the previous training time.