摘要
7 最大熵方法80年代初人们开始考虑用信息论中熵的概念进行阈值选取。1980年Pun首先提出了最大后验熵上限法。两个后验熵定义为H_B^(?)=-P_tlog_2P_t和H_w^(?)=-(1-P_t)log_2(1-P_t)。
Thresholding is a well-known technique for image segmentation in digital image processing. It finds wide applications in various areas, such as machine vision, character recognition, biomedical image analysis, fingerprint and seal identification, optical stripe discrimination and military target recognition. Threshold selection is crucial to thresholding. This paper more completely summarize more than fifty methods for threshlod selection proposed and developed in succession in the last thirty years (1962-1992) in accordance with a large number of the relevant literatures, and consequently these methods are classified into ten categories: (1) histogram methods and histogram transformation methods; (2) maximum between-class variance method; (3) minimum error method and uniform error method; (4)co-occurrence matrix methods; (5) moment-preserving method; (6) maximum entropy methods; (7) simple statistic method and local feature method ; (8) probability relaxation method; (9) fuzzy set methods; (10) other methods. The basic ideas of the various methods as well as their advantages and defects are critically reviewed. Some corresponding formulae of computation are also given. Some limitation in evaluating the performance of these methods quantitatively are pointed out.
出处
《数据采集与处理》
CSCD
1993年第4期268-282,共15页
Journal of Data Acquisition and Processing
关键词
图像处理
阈值选取
image processing, threshlod value, thresholding, histograms, maximum entropy, probability relaxation, fuzzy sets
