Image segmentation method based on validity index of fuzzy clustering

Abstract

The invention discloses an image segmentation method based on a validity index of fuzzy clustering, which comprises the steps of 1, performing classification by using fuzzy C-mean clustering algorithm; 2, building an objective function, and judging whether a termination condition is met or not or whether the maximum number of iterations is reached or not; 3, performing initialization and updating a membership degree matrix and a clustering center; 4, calculating the compactness and the separability, and acquiring an index value; and 4, acquiring the optimal clustering number at a maximum value of the index value. The image segmentation method can perform accurate division on a pixel set and is applicable to complex and overlapped pixel sets with noise pixels, thereby being capable of performing good segmentation on an image.

Description

technical field [0001] The invention belongs to the field of data mining, in particular to an image segmentation method based on the validity index of fuzzy clustering. Background technique [0002] Fuzzy C-means clustering is the most widely used algorithm in fuzzy clustering, and a lot of research has been done on this algorithm. It is known through a lot of research that no clustering method can obtain the optimal division of all sets; secondly, many clustering algorithms need to input the number of clusters in advance, but usually the optimal number of divisions of a data set before clustering is Unknown. The process of obtaining the optimal number of clusters through the clustering validity index is an iterative process. By continuously changing the initial value c of different clusters, the corresponding validity index value of each division is calculated, and finally the obtained index is analyzed and compared. The size and change of the value, usually the highest v...

AI Technical Summary

Problems solved by technology

In 1998, Rezaee proposed to scale the compactness and separation in the XB index through the proportional factor, and obtain the new index by linear combination. Although the overall performance has been improved, the structure of the index is complicated, and there will be problems with Judgment that deviates from the facts, unstable performance

Since then, some people have continued to improve and perfect the indicator, but the improved indicator is too complicated and the calculation efficiency becomes low

In 2004, Pakhira et al. proposed the PBMF index, which is an effective index that can evaluate the results of hard clustering and fuzzy clustering at the same time. The index consists of three factors. When the number of classes is small, the second and second The three factors play a larger role. When the number of classes increases, the role of the first factor becomes more and more obvious. This indicator does not fully consider the overlap of data sets.

In 2011, H.Le Capitaine et al. proposed the OSI indicator, which uses multiple clustering and separation methods for the measurement of each data point, and is based on the degree of aggregation of members. Although this indicator takes into account the overlap between data sets, However, the calculation method is too complex and cumbersome, which makes the time complexity and space complexity large

In 2015, Chih-Hung Wu and others proposed the WLI index, which solved the shortcomings of existing indexes, but it could not achieve good results for data sets with complex data structures and large and small clusters; in 2016, Zhao Nana, Qian Xuezhong proposed the effectiveness index CSO based on compactness, overlap, and separation. This data set realizes the judgment of the optimal clustering number of data sets with overlapping subclasses between classes, but this index is not suitable for massive high Dimensional datasets and datasets with special shapes did not achieve good results

Images