| Summary: | This study addresses the problem of gender classification using frontal images. We have developed a gender classification with performance superior to existing gender classifiers. The first step is that the face image is projected into a face space via Principal Component Analysis (PCA) to reduce dimension. And then this face space is projected onto LDA vector to construct a classifier. We separate the face data into different training groups, and derive different numbers of Principal components (20 and 40 components). Comparing the results, the group using the most training images with the larger numbers of components, 40-components, yielded the best accuracy rate 92.9%.
|
|---|