Image Compression - Bezier and Splines in Image Processing and Machine Vision - page 98

higher. Other components of the fidelity vector are practically the same. This
is also true for the reconstructions based on equispaced surface pixels. All the
reconstructed images have different values in entropy from that of the input
image. This change is due to merging of small regions in the segmentation
procedure before encoding of the input image and polynomial approximation
in the reconstruction process. Fractal dimension of the reconstructed images
differ slightly from that of the input. This is probably due to the reason that
contours of the reconstructed images are not as smooth as that of the input.
Below in Figure 4.4, we present reconstructed Lincoln image along with the
input for visual comparison.

(a) (b)

(c)

Fig. 4.4. Reconstruction of Lincoln image: (a) input Lincoln image; (b) reconstruc-
tion from coecients; (c) reconstruction from surface points.

4.4.2 Results of SLIC Algorithm for 256 X 256 Images

We now discuss the results of the compression algorithm on two famous 8-bit
images (Lena and Girl) where each one is of size 256 × 256. These two images

are more complicated than the previous 64 64 Lincoln image, because these

images have texture regions in them and the texture blocks as seen from Table
4.3 have taken a considerably large number of bits, lowering the compression
ratio. The number of contour pixels on line and arc segments for the Lena

×