THE FEATURE MECHANISMS IN THE VISUAL PATTERN RECOGNITION PROCESS

This study had two primary objectives. First, this experiment investigated the process of pattern recognition--how did human observers recognize a presented pattern stimulus? And how did the contemporary models--for example, the descriptive and feature process models, and signal detectability theory--apply to the current confusion data? Second, the experiment then focused on the mechanisms of feature processing (sampling) within the framework of a feature-analysis approach and signal detection analysis to the visual pattern recognition problem. The present experiment consisted of complete identification of a set of stimulus patterns constructed from the set of all possible combinations of three equal-length line features. Two variables were manipulated: one was the stimulus exposure duration; the other was the payoff structure. A common type of feature-analytic pattern recognition model is comprised of a sensory sampling process followed by an independent decision process. It was assumed that stimulus exposure duration would primarily affect the early process stage and that payoff structure would basically affect the later process stage. The results rejected all descriptive models as well as the Global-to-Local model but not the Similarity-Choice model and falsified every basic assumption in the feature process models. In addition, the finding demonstrated the potential of statistical dependency among feature channels at the feature sampling probability level as well as the d\u27 and (beta) in the signal detection analysis. Signal detection analysis was used to help explain the perceptual effects of stimulus and response aspects in different experimental conditions. The overall averaged marginal d\u27 (feature senstitivity) remained relatively constant as more features were contained in a stimulus, while the marginal (beta) (feature decision bias) showed a decreasing trend pattern. By analogy, the conditional d\u27 and (beta) microscopically illustrated the evidence of positive interdependencies in both feature sensitivity and decision bias. Finally, the results also revealed that both independent variables did affect the corresponding parameters in the model structure