Portraits and perception: configural information in creating and recognizing face images
Portraits and perception: configural information in creating and recognizing face images
复制标题
DOI：
10.1163/156856808782713843
复制
发表时间：
2008-01-01
期刊：
影响因子：
--
通讯作者：
Sinha, Pawan
中科院分区：
文献类型：
--
作者：
Balas, Benjamin J.;Sinha, Pawan
长期以来，配置信息一直被认为对于人脸识别很重要。然而，传统的肖像画教学鼓励艺术家对面部使用“通用”配置，而不是试图复制精确的特征位置。我们通过两项任务来研究这个有趣的悖论，这些任务旨在测试配置信息融入面部表征的程度。在实验 1 中，我们使用简化的面部生成任务来检查如何准确地将特征配置合并到生成的相似性中。在实验 2 中，我们询问实验 I 中创建的“肖像”是否可以与真实图像区分开。这些实验的产生和识别结果显示出一致的模式。受试者不太擅长将面部特征（眼睛、鼻子和嘴巴）排列在正确的位置，以及区分错误配置和正确配置。这种对构形关系看似不敏感的做法与艺术家基于通用几何模板创作肖像的实践是一致的。有趣的是，艺术家隐含地使用这个通用模板的参考框架 - 外部面部轮廓 - 在我们的实验结果中成为性能的重要调节器。在存在外部轮廓的情况下，可以减少生产错误并提高识别性能。我们讨论了这些结果对人脸识别模型的影响，以及为什么肖像如此难以创建的一些可能的感知原因。
Configural information has long been considered important for face recognition. However, traditional portraiture instruction encourages the artist to use a 'generic' configuration for faces rather than attempting to replicate precise feature positions. We examine this intriguing paradox with two tasks designed to test the extent to which configural information is incorporated into face representations. In Experiment 1, we use a simplified face production task to examine how accurately feature configuration can be incorporated in the generated likenesses. In Experiment 2, we ask if the 'portraits' created in Experiment I are discriminable from veridical images. The production and recognition results from these experiments show a consistent pattern. Subjects are quite poor at arranging facial features (eyes, nose and mouth) in their correct locations, and at distinguishing erroneous configurations from correct ones. This seeming insensitivity to configural relations is consistent with artists' practice of creating portraits based on a generic geometric template. Interestingly, the frame of reference artists implicitly use for this generic template - the external face contour - emerges as a significant modulator of performance in our experimental results. Production errors are reduced and recognition performance is enhanced in the presence of outer contours. We discuss the implications of these results for face recognition models, as well as some possible perceptual reasons why portraits are so difficult to create.