| OCR Text |
Show "intermediate" and "worst" images, respectively, and the segmentation parameters obtained by manually optimizing the segmentation quality of these frames were used to process the remaining frames in the image database. Figure 5.64(b) presents the comparison between the adaptive technique and the traditional approach using the "intermediate" and "worst" images. The average segmentation quality for the traditional approach using the "intermediate" and "worst" images was 74.7% and 75.1%, respectively. Even though these images lead to better performance than the traditional approach using the first image, the adaptive segmentation technique is still superior to the traditional approach. As Figure 5.64 shows, the performance of the default parameter approach and the traditional approach was highly erratic throughout the sequence of the outdoor images. Figure 5.65 illustrates the quality of the segmentation results associated with the default parameters and the traditional approach using the first image in the outdoor image database and contrasts this performance with the adaptive segmentation technique. Each result corresponds to the average segmentation performance produced by each technique for Frames 1 and 11 o f the database. The left and right columns o f this figure present the segmentation results o f Frames 1 and 11, respectively. By comparing the extracted car region in each of these images, as well as the overall segmentation o f the entire image, it is clear that the adaptive segmentation results are superior to the other methods. 5.2.3 Sequential Testing Experiments The outdoor experiments described above were conducted in a parallel fashion, i.e., all training and all testing were performed without the aid o f previous segmentation experience obtained during their own phase. Although the testing experiments used the knowledge acquired during training, the tests were still performed in parallel. None o f the segmentation experience obtained during testing was applied to subsequent testing images. 165 To measure the improvement in efficiency achieved by immediately reusing segmentation experience, I conducted a final set of experiments on the outdoor image |
