In this example, the gray-scale cameraman image is used to obtain the principal components (eigenimages). These eigenimages are used as basis functions for compressing the image. The image is partitioned into 8 8 block and the 64 64 sample covariance matrix is formed. There are 64 basis functions. Each basis function is an 8 8 image. To compress the image, we use 16 of these principal component images corresponding to the 16 largest singular values. The original image, the basis functions, the compressed image, and the error image are shown in Figures 3.19 through 3.22, respectively. The peak signal to noise ratio (PSNR) of the compressed image is 65 dB. Note that the basis functions are not quantized. If we quantize the basis

functions and the weights of the basis function, the PSNR will decrease and the compression losswill be more visible.

Transcribed Image Text:

FIGURE 3.19 Original image. FIGURE 3.20 Eigenimages.