Quantifying the effects of content, complexity and delay on memory for real-world images

“A picture speaks a thousand words”, yet retrieval of a specific, previously viewed image from a large image collection may result in a substantial number of unwanted returns. This thesis offers an investigation into the feasibility of a novel solution to the retrieval of pictures. Rather than offering an alternative to current image retrieval strategies, research presented examines whether a novel approach, grounded in human cognition, can have specific benefits within specific contexts. The HELM analytical model (Lansdale, 1998) was adapted to 8 experiments observing location memory for everyday pictures. Experiments 1 and 2 replicated some of the earlier work conducted by Lansdale, Oliff & Baguley (2005), and examined memory for location in a novel set of real-world images. Experiment 3 looked at the impact of image content on the distribution of errors in recall. Experiment 4 examined the effect of threat-victim relationships between objects on memory for location. Experiments 5 and 6 were concerned with the effects of object density within a scene. And Experiments 7 and 8 examined the effects of forgetting of spatial memories over periods of recall delay. Data from these experiments demonstrated that memory for location in complex real-world images is prone to two types of error, near-miss errors which fall at locations neighbouring the target location, and far-miss errors which fall at locations distant to the correct location value. The presence of threat within an image resulted in participants foreshortening the estimation of space between two objects. Recall was shown to be more accurate for single-object over multiple-object images. And forgetting can be characterised as both a loss of the availability of location information in memory, and a spread of the distribution of error in recall. Findings are discussed in terms of the applied consequences for the design of query languages for image retrieval.