Face processing
Claus‐Christian Carbon · Refubium (Universitätsbibliothek der Freien Universität Berlin) · 2003
In the present work, early processes in the recognition of faces are investigated. Short presentation times of 200 ms were chosen because within this time window the noise in the behavioral data which is produced by eye saccades is reduced to the minimum. Moreover, underlying perceptual recognition strategies seem to be less cognitive penetrable (cf. Pylyshyn, 1999) within such a short time window. The work is separated in two distinct empirical parts. The first part (Experiment 1 and Exp.2), deals with unfamiliar faces and the processing order of several face features The second part (Exp. 3a, 3b, 4, 5) uses familiar faces and explores the role of local vs. configural and holistic facial information in the recognition of faces. In two pre-studies the facial material for the first experimental series was constructed. The artificial faces consisted of parts of natural faces. Based on these natural faces, the eyes-, nose- and mouth-region were systematically manipulated. This was done in a high salient and a low salient way. Moreover,locally and configurally changed variants were constructed. In the local version, the focus points (eyes, nose, mouth) were replaced with alternative features. In the configural version, the eyes were shifted inwards, and the noses and the mouths were shifted downwards. In order to test specific feature processing models, a stimulus limitation paradigm with backward masking was used. The participants had to answer quickly as well as accurately, whether two sequentially presented faces were same or different. The presentation times (PTs) of the target face (second face) were varied between 32 ms and 94 ms. Then, the percentage of correct rates were calculated for every combination of feature manipulations (E: eyes, N: nose, M: mouth, EN, EM, NM, ENM), every PT and both=0Dmanipulation classes (local vs. configural). The results show that the data-fit is best for two different processing models with respect to the two used manipulation classes. First, analyzing the data for local changes, a strict-serial microgenetic model seems to be the most preferable processing order. Thus, participants processed changes to faces, which were locally changed, in a serial way. Starting with the recognition of the eyes, the further recognition processing mentally jumps to the mouth area. After having processed the mouth, the nose will be processed. Second, a totally different processing pattern seems to be pursuit, if the participants had to detect changes of configurally changed faces. Here, they processed the faces in a holistic way. With a little priority for the eyes detection, all succeeding features were processed in parallel. Therefore, local and configural changes seem to be dissociable on the basis of the underlying recognition processes. The second experimental series investigated the role of local identification processes within the early recognition of faces. In order to test, whether local identification processes are beneficial for the identification of a face inverted Thatcher-faces (Thompson, 1980) are used. In such faces, the eyes- the mouth-regions are turned upside-down. Interestingly, these tremendous changes are hardly detectable when the faces are inverted. Therefore, Thatcher-faces seem to be ideally suitable for testing the role of local identification, because the only changing to these faces is the turning of these two discrete regions, without changing the overall appearance of the stimuli at all. If inverted Thatcher-faces will be faster recognized than inverted normal faces, then the RT-advantage must be caused by the specific manipulation of the Thatcher-faces. The eyes- and mouth regions of inverted Thatcher-faces are already in a correct orientation, but this is not the case for inverted normal faces, where these regions are turned upside-down. Thus, the RT-benefit would be caused by the identification of local structures, which have not to be rotated according to the mental rotation hypothesis (Shepard & Metzler, 1971) as their counterparts of the inverted normal faces have to. The alternative hypothesis assumes that Thatcher faces will be recognized with the same speed or even slower than normal faces. The same speed would be assumed only if the outline of faces can be recognized in the used PTs of only 26 or 200 ms, respectively.Longer RTs would be assumed if holistic or template-like recognition processes are responsible for the early processing of faces. Experiment 3 and Experiment 5 revealed that inverted Thatcher faces were processed faster than inverted normal faces when they were presented for only 26 ms. Thus, the hypothesis of early identification processes was supported. However, using a PT of 200 ms, this RT-relationship was inverted. Now, the recognition of complete, i.e. normal faces was faster than the recognition of Thatcher-faces. Therefore, it seems that only with very limited time resources, local identification processes are particular beneficial for the recognition of faces. If there are no such time constraints, then holistic face processing strategies seem to be more advantageous. The present work uses important new experimental paradigms to test specific face processing models. It demonstrates not only that local face information plays a distinct role for the recognition of faces under different time resources. Moreover, it shows that local and configural information is processed in a different way.