Analytical and empirical methods were proposed for the evaluation of segmentation algorithms by Zhang (1996). The later
indirectly judge the algorithm by evaluating the quality of segmentation result. In the case that we have a priori knowledge of the correct or approximately correct segmentation given by the experts, segmentation evaluation usually can be made by visual comparison between the initial image and the segmented image (Martin et al., 2006). Udupa et
al. (2006) stressed that the efficacy of a segmentation algorithm in an application is to be measured in terms of three
factors: precision, accuracy and efficiency. In comparison of a segmentation algorithm to a ground-truth segmentation
of the image, Bowyer (2000) argued that five possible outcomes need to be identified: correct segmentation, oversegmentation, under-segmentation, missing a region and an incorrect segmentation of a noise region. Therefore, we also evaluate segmentation algorithms of beef-marbling images from the aspects in this study. Our approach for the evaluation of segmentation consists of the three steps. Firstly, panelists were trained to evaluate the algorithms performance. Secondly, we used surrogates of truth established by manually thresholding as references for the comparison of segmentation methods since it is impossible to establish absolute true segmentation. Finally, supervised evaluation method known as relative evaluation method was employed to evaluate segmentation efficacy by comparing the resulting segmented image against amanually segmented reference image, which is often referred to ground truth (Zhang, 1996). Based on the degree of similarity between the human and machine segmented images, the panel determined the quality of the segmented images with a six-point scale: 5 = the segmented beef-marbling image completely agrees with its corresponding reference, 4 = the segmented beef-marbling image is similarly agreement with its corresponding reference, 3 = there is the over-segmentation or under-segmentation in the segmented beef-marbling image, 2 = the segmented beefmarbling image visibly differs its corresponding reference,
Analytical and empirical methods were proposed for the evaluation of segmentation algorithms by Zhang (1996). The laterindirectly judge the algorithm by evaluating the quality of segmentation result. In the case that we have a priori knowledge of the correct or approximately correct segmentation given by the experts, segmentation evaluation usually can be made by visual comparison between the initial image and the segmented image (Martin et al., 2006). Udupa etal. (2006) stressed that the efficacy of a segmentation algorithm in an application is to be measured in terms of threefactors: precision, accuracy and efficiency. In comparison of a segmentation algorithm to a ground-truth segmentationof the image, Bowyer (2000) argued that five possible outcomes need to be identified: correct segmentation, oversegmentation, under-segmentation, missing a region and an incorrect segmentation of a noise region. Therefore, we also evaluate segmentation algorithms of beef-marbling images from the aspects in this study. Our approach for the evaluation of segmentation consists of the three steps. Firstly, panelists were trained to evaluate the algorithms performance. Secondly, we used surrogates of truth established by manually thresholding as references for the comparison of segmentation methods since it is impossible to establish absolute true segmentation. Finally, supervised evaluation method known as relative evaluation method was employed to evaluate segmentation efficacy by comparing the resulting segmented image against amanually segmented reference image, which is often referred to ground truth (Zhang, 1996). Based on the degree of similarity between the human and machine segmented images, the panel determined the quality of the segmented images with a six-point scale: 5 = the segmented beef-marbling image completely agrees with its corresponding reference, 4 = the segmented beef-marbling image is similarly agreement with its corresponding reference, 3 = there is the over-segmentation or under-segmentation in the segmented beef-marbling image, 2 = the segmented beefmarbling image visibly differs its corresponding reference,
การแปล กรุณารอสักครู่..
