Examining Spatial Properties From Multiple Views In Scene Recognition

Our spatial representation of the environment allows us to efficiently and appropriately use spatial information to identify objects or scenes, and to control our behaviors. Such spatial representation should be flexible and dynamic as people rarely experience the world from a single viewpoint because objects in the environment and the observers often move around. Extensive researches on mental rotation have explored how people recognize objects from a different viewpoint following the initial exposure, that is, how human transform the mental representation of the initial object/scene to a new viewpoint. Thus, the mechanisms which underlie how people represent the spatial property information to facilitate them to recognize the environment have caused the heated discussion.Two groups of theories have been proposed about objects recognition: view-dependent theory and view-independent theory. Recognition performance generally supports viewpoint independent theory by presenting objects that are composed of distinctive, identifiable components and supports viewpoint dependent theory, when the stimuli are not readily distinguishable or labeled by a definite name, such as wire-frame objects and blob-like objects. Although in object recognition, the feature of the components have received much attention, in scene recognition, most research only focused on the scenes that contained unique objects. Since the environment that we perceived everyday is quite complicated and full of unique objects, it is conceivable that when all kinds of the information are available in an environment, the observers might encode or process different features (e.g. object position and identity) in different ways.We explored above issues through a spatial task involving changes in viewpoint in a real scene with the visual change-detection paradigm. The program designed to develop the templates was created using Matlab 7.0.4. This program generated the layouts by randomly determining positions for each of the objects on the table. The object arrays were comprised of five or seven same glasses or various types of cups. Participants were fitted with a blindfold and were instructed on how to respond using the microphone. There were 38 volunteers to take part in the research, who were tested individually. A 2 (set size: 5/7)×2 (viewpoint: 0°/50°)×5(P-Identical, P-Similar, P-Different, I-Similar, I-Different) within subject design was conducted and gender was set as a between subject variable in Exp 1. Position task asked subjects to report which object was moved and the identity task need them to report which one was replaced. The learned phase lasted for 5s and the rehearsal phase lasted for 10s. The reaction time (RT) was calculated on the correct responses with its log values. Error rate was defined as the number of wrong responses out of a total of 16 trials for each experimental condition. The results firstly indicated that identity information of objects was matched to identity task, which could eliminate the advantage of position processing that happened in previous researches. When viewpoint change happened, the performance of position task decreased sharply, while no obvious influence existed on performance of identity task. As the number of objects increased, the performance of identity information decreased in a large extent, while the performance of position information probably kept a stable condition. Male showed a kind of advantage on mental rotation task, while female seems to do better works on identity task.According to the result of Exp1, the verbal encoding played an important role when participants complete the spatial property tasks. Therefore, we replaced the variable of set size of objects by the variable of verbal interference in order to further explore the effect of verbal encoding. A similar 2×2×5 experimental design was conducted in Exp 2. On the one hand, the result of Exp2 was consistent with Exp1 on the relation between viewpoint and spatial property processing. On the other hand, the result also suggested that verbal interference had a larger influence on identity task than on position task. Meanwhile, the effect of viewpoint change on scene recognition was more sufficient than the effect of verbal interference. Moreover, when the identity information of subjects in the spatial scene was not apt to be named by verbal encoding, the influence of verbal interference would be the most significant of all conditions both in position task and identity task.The current research expects to explore the inherent mechanism of representation of spatial scene by combining several experimental paradigms used by former researchers in various kinds of settings. We execute the procedure of experiments in daily three-dimensional spatial scene in order to avoid the unilateralism of viewpoint provided by computer screen. The results could be concluded as followed: the performance of scene recognition was relative to the match between spatial property information and task, which implied that the process of spatial property information was probably task-oriented. The position representation might be view-dependent and the identity representation might be view-independent during scene recognition. The verbal encoding could facilitate the identity task and observers were likely to process the identity information automatically, although the effect of viewpoint change was more sufficient.