基于灰关联分析和最大熵阈值的医学彩色图像分割算法

An Algorithm of Medical Color Image Segmentation More Effective than Existing Ones in China

在彩色图像分割领域,因色彩信息丢失而导致的分割精度差和因运算量大而导致的实时性不好是两大亟待解决的问题。文中将灰关联分析理论引入彩色图像分割领域,提出一种基于灰关联分析和最大熵阈值的医学彩色图像分割算法。避免了传统彩色图像分割中因三维色彩信息发散而造成的信息丢失现象,提高了分割精度,又能提高分割实时性。实验结果证明了该算法的有效性。

Aim. To our knowledge and in our opinion, Ref. 2 by Liang et al deals with medical color image segmentation more effectively than other Chinese papers. We now propose a medical color image segmentation algorithm which we believe is more effective than that of Ref. 2. In the full paper, we explain our algorithm and its effectiveness in some detail~ in this abstract, we just add some pertinent remarks to listing the three topics of explanation. The first topic is： the principles and procedure of the algorithm. In this topic, we transform the RGB （Red, Green, Blue） color information of a medical image into array vectors as comparative sequences and take {1,1,1} as reference sequence. Then we calculate their grey relational coefficients and grey relational degrees, as given in eq. （1） and eq. （2） in the full paper. The calculation results produce grey relational images, which are further segmented by using the maximum entropy thresholding method. The second topic is. experiments and the analysis of their results. In this topic, we do experiments on the color image segmentation of the two images respectively of two cerebrum slices respectively taken from two different positions; the segmentation results are shown in Fig. 1 and Fig. 2. We also compare the segmentation results of our algorithm with those of the algorithm contained in Ref. 2. The comparison results reveal that the segmentation error rate of our algorithm is 1. 34% ,lowerthe error of the sum of all edges is 0.28% less; the computing time is 1/10 that of the algorithm in Ref. 2. The third topic is： quantitative evaluation and conclusions. In this topic, we define segmentation quality evaluation （SQE） and segmentation effectiveness evaluation （SEE）; the SQE results, given in Table 1, show that the segmentation quality of our algorithm is about 6 times better than that of Ref. 2. Then we use the SEE to compare the effectiveness of our algorithm with that of the algorithm in Ref. 2; the comparison results, given in Table. 2, indicate that the SEE of our algorithm is 2.43 times the best of 3 SEEs of Ref. 2＇s algorithm.