Research On Visual Gender Processing

Gender information is quite important in our daily life. Human beings face different kinds of objects and people various from age, gender and race. When facing with each object, which is gender-discriminable, like face, hand, body, cloth, shoes, even some language character [40] [9], most of us could make the gender-decision quickly without few exceptions at early age of our life. In consideration of series of behavioral experiments, daily observation and clinical outcome, the influential Bruce and Young model [42] of face processing proposed that different facial aspects are handled by different specialist processing subsystem. The process of deciding whether a face is male or female (referred to in the literature as either sex decision or gender decision) was assigned to a component labeled 'directed visual processing'.Moreover, in all probability, there exists a gender-processing unit, which is in charge of all the tasks related to gender. Therefore, an experiment involved with objects, which contain explicit gender information like shoes and clothes, instead of facial gender could be investigated.On the other aspect, an event-related potentail (ERP), which is a measure of the brain's response to a sensory stimulus, is measured with EEG. The ERPs are very samll in comparison with the ongoing EEG and are barely visible in an individual trial. Analysis of ERP relies on the identification of signals after averaging serveral presentations of the same stimulus patterns. However, this method ignores the fact that the response may vary widely across trials in amplitude, time course, and scalp distribution.Thus, in this paper, with these two problems above, we applied ERSP (Event-Related Spectrum Permutaion) method and designed method class GRSEC (Gender Related Single-Trial EEG Classifier) to our datasets. The results indicate that there exists significant difference between the response invoked by two kinds of stimuli, male and female. Althought the facial gender's response is greater than the object's, the similarity between them is obvious; The designed GRSEC performs an average accuracy of 71.66%(f-measure=0.71), which proved that it can handle the single-trial classification problem.