The Choice of Quality Indicator for the Image Segmentation Evaluation

Abstract

In this paper, the main attention is focused on the stage of image segmentation and on the choice of a quality indicator for the evaluation of image segmentation. In general form the segmentation problem of color and tone images is formulated in. The main approaches that are used in methods and techniques of image segmentation are highlighted. The need to evaluate the results of the work of methods and techniques of image segmentation to assess their performance has been established. The options for the image segmentation quality assessing of both at the objective (quantitative) level and at the subjective (qualitative) level are considered. The main features of qualitative segmentation are highlighted. Two main groups of indicators for evaluating the quality of segmentation were analyzed: indicators based on comparison with the reference result of image segmentation and the indicators for which benchmark segmentation is not required. Their main disadvantages when used in evaluating the results of image segmentation are described. The information quality indicator – Kullback-Leibner divergence for the image segmentation quality assessing of the results of image segmentation is considered. The result of image segmentation by the Canny edge detector is presented. The estimation of the quality of the segmented image was calculated using the information Kullback-Leibner divergence for pairs of images of different scales. The paper shows show a graph of the dependence of the value of the Kullback-Leibler divergence on the value of the scale factor of the original image. Keywords— Canny edge detector, image segmentation, objective criteria, Kullback-Leibler divergence, quality indicator.