| Summary: | 碩士 === 雲林科技大學 === 電子與資訊工程研究所 === 97 === Face recognition is an important issue in pattern recognition. Linear discriminant analysis (LDA) has been widely used in face recognition. However, the small sample size (SSS) problem occurs when the number of samples is far smaller than the dimensionality of the image, which makes it not so representative and leads to the problem of the within-class scatter matrix. What’s more, it may lay an emphasis on the effect of the recognition. Therefore, we propos a modified LDA (called block LDA) in this study to divide the input image into several non-overlapping subimages with same size. With the proposed method the quantity of the samples is increased and the dimensions of the sample space is reduced. In addition, to reduce the impact on illumination variations, the input image was transformed into gradient image. The experimental results show that this method indeed solves the SSS problem with a good recognition rate.
|
|---|
