Download Now Free registration required
In this paper, face recognition technique using Discrete Cosine Transform (DCT) is proposed. The local information of the face is extracted using block-based (DCT). The coefficients selected in each DCT block are fused to generate the feature image. This feature image is used for classification process. The face, recognition is then performed using Mahalanobis distance. The advantage of this technique is the reduction in the dimension of the face space retaining low, mid and high frequency coefficients. The technique is validated using standard ORL and YALE face datasets. The experimental results outperform traditional methods like PCA, LDA and DCT normalization.
- Format: PDF
- Size: 461.9 KB