Figure 1: (a),(b) and (c) are recognized 15ms faster than the general mean. (d),(e) and (f) show no significant global pattern. According to pairwise Sheffe post-hoc test, recognizing (f) takes 20ms longer than recognizing a similar blue image (e)
Motivation and MethodWater is fluid and changeable, its colour and shape are determined by its environment. For example, deep and shallow water require very different physical simulations [Darles et al. 2011]. Yet humans effortlessly discriminate between water and non-water: there must be simple and unique perceptual features making discrimination possible. We are conducting an empirical study of water recognition and here report the preliminary results.In a series of experiments test images of water (photographs, realistic drawings and abstract depictions), are paired with similar non-water images. Volunteer participants see a pair of images side by side and as quickly as possible press a key corresponding to the image that represents water. In some experiments the image pair is displayed until a response occurs, immediately followed by the next pair. This sequence confused subjects because of temporal binding between the key press and the next pair of images. A brief grey mask was added immediately after each response, but an afterimage of the previous pair persisted. We replaced it by randomly generated coloured masks for the remainder of the experiments to prevent previous colours and edges from interfering with the next trial. Yet, the task was too easy, the resulting data showed little statistical significance. This improved with image presentation shortened to 150ms followed by a coloured mask, which was displayed until a selection was made. We recorded error rates and response times of 12 participants, which we analyzed using ANOVA. Each participant's data was normalized to compensate for regression. Outliers were recursively removed [Selst and Jolicoeur 1994]. To exclude the effect of the subject we analyzed distances of the reaction times to the mean. In some experiments we used all pairs of images including water/water and distractor/distractor, with subjects instructed to guess if both images were equally waterlike. The average response time for valid trials with a test and a distractor was 460ms.; trials with two distractors took 144.7ms. longer. The effect is consistent with the previously observed differences between target and non-target visual stimuli [Johnson and Olshausen 2003]. Thus we eliminated such trials. Some distractors made the subjects slow down or make mistakes: periodic functions, spirals, horizontal lines, uniform grey textures and blue sky. Such distractors contain simple visual features commonly found in water, which are registered preattentively. We tried to use dissimilar distractors to isolate the effects of the tests. For example, sky is often seen reflected in water and unsurprisingly accounts for twice the mistakes. In contrast, contextually unrelated blue distractors did not have a significant effect.
ResultsAll easily recognized ...