| OCR Text |
Show I example, the four points located at the bottom plateau of the surfaces (both the global and local segmentation quality 0.0) in Figures 7.5(a) and 7.5(b) are plotted at the same lower left comer in Figure 7.7(a). Figure 7.7(c) displays the utopian point at the upper right comer, which caused the termination of the genetic algorithm after the third generation. Figure 7.8 displays the performance curves that indicate the maximum and average fitness values of both the global and local quality measures of the population during each generation for the representative images. Maximum fitness values continuously increase in these charts because the highest fit individual for either the global or local measure is always retained from one generation to the next. Average fitness values, on the other hand, fluctuate as the individuals visit different regions of the surfaces in search of highly fit areas. However, they increase gradually as the genetic search process progresses. To provide a visual indication of the performance improvements achieved by the adaptive image segmentation system for multiobjective optimization, Figures 7.9 and 7.10 show the initial and final segmentation results for these representative frames. The segmentation results shown in these figures were obtained from the individuals in the short-term population with maximum global fitness (e.g., the best global segmentation quality) or maximum local fitness (e.g., the best local segmentation quality). The image regions extracted by the segmentation process are shown as pseudo-color images with the boundaries superimposed on them. An increase in overall segmentation quality between the initial and final results can be seen in these figures. The global segmentation results, which optimized the segmentation quality measure of the whole image, show a trend to obtain more precise boundary representations for all objects including the car in each image. In the local segmentation results which optimized the segmentation quality measure of the object regions of interest, the portion of the car that is extracted from the images becomes larger in the final results. Notice that the bottom of the car in the outdoor image is extracted as a separate region from the background in the final result, although this region is still not combined with the top portion of the car to form a single region. 1 1 1 |
