Human expression study based on bubble technology.
详细信息
文摘
Humans are able to recognize facial expressions of emotion from faces displaying a large set of confounding variables, including age, gender, ethnicity and other factors. This involves categorization of various information from the same structure. Much work has been dedicated to attempts to characterize the process by which this highly developed capacity functions. In this thesis, we propose to investigate local expression-driven features important to distinguishing facial expressions using a so-called Bubbles technique [4]. The bubble technique is a kind of Gaussian masking to reveal information contributing to human perceptual categorization. Gaussian filters are randomly distributed in area to cover part of the whole information space. Observers are required to browse through the bubble-masked expression image and identify its expression category. By collecting responses from observers and analyzing them statistically we can find the facial features that humans employ for identifying different expressions. Humans appear to extract and use localized information specific to each expression for recognition. To conduct 2D bubble-based expression recognition, we collected facial expression images of three male and three female from our BU-4DFE database as the test samples. Display sequence is carefully designed to assure an exhaustive and exclusive experiment. Each trail is an expression sample filtered by the bubble mask with two choices. Observers choice and the bubble mask are saved. Two kinds of experiments have been conducted. First one includes 78 observers and each goes though 72 trials. The other one includes 17 observers but each goes though 432 trials. Then the classification image is generated by adding valid bubble mask together and convoluting the summation with Gaussian window. Finally the pixel test and cluster test search the smoothed classification image for significant signals that help humans make the correct judgment. The key features humans use to identify different expressions should reside within these diagnostic regions. To further excavate the information space, we have also investigated an approach for 3D bubble face generation. A GPU pipeline is used to render the bubble masked 3D face model to achieve good visual effect and liberate CPU from the major graphic computation so it can work more efficiently on the data processing stage. A vertex shader and a fragment shader are provided to draw each pixel on the model. The connection between general programming language C＋＋ and specific graphic programming language Cg has also been built. By implementing multiple texturing and GPU-based blending technologies, vivid bubble effect can be created in the 3D domain. The major contribution of this work lies in tow-fold. First, we proposed a novel idea to investigate the feature regions of expressions in a certain frequency utilizing the new bubble-based image masking approach. Second, we extend the 2D bubble generation approach to our 3D facial models, resulting in a special effect of bubble-masking on them. We propose a new implementation strategy by using both GPU and CPU in parallel to improve not only the efficiency of bubble image creation but also the efficacy of experimental execution.