本研究之目的爲發展適用於磁振影像之自動影像分割方法,以分析水果內部損傷區域並估算其損傷體積。所選擇的實驗對象爲加州李與國產水蜜桃兩種水果,水果經過不同程度之撞擊後以磁振造影儀取像,再分別利用以非監督式影像二元化方法爲核心的影像分割演算法分割出影像中水果內部損傷的區域,從而估算截面積與損傷體積。我們測試了四種影像二元化方法的處理效果,所估算得之內部損傷體積分別與人工判斷影像分割所得之結果進行比較。實驗結果顯示以加州李做爲檢測之樣本,四種分割方法的體積估算相對平均誤差介於9.4~20.2%之間。對水蜜桃磁振影像的體積估算相對平均誤差較大,介於14.8~26.4%之間。以影像分割的穩定度指標與體積估算誤差為基礎比較四種影像二元化方法的效能,四種方法中除了Kapur et al.法較不適用外,其他三種方法均可以適用於影像分割的二元化處理,而三種方法中又以保矩法之效能較好。利用本研究所發展的影像分割方法進行水果內部損傷區域的分析,除了可以較客觀地進行水果內部損傷體積的估算,也大幅地減少了人工處理磁振影像之時間與人力。
The objective of this research is to develop a feasible image segmentation method for bruise volume estimated from magnetic resonance images of selected fruits. Plums and peaches were subjected to various levels of impact and their magnetic resonance images were acquired with a magnetic resonance imaging system. The internal bruise regions were segmented with the developed method that was based on an unsupervised thresholding scheme. The bruise volumes were then estimated by integrating bruise regions in multiple image slices. Four unsupervised thresholding schemes were tested and evaluated by comparing the estimated bruise volumes with the manually estimated bruise volume. For bruise volume estimation of plums, the average relative error ranged from 9.4% to 20.2% while the average relative error for peaches ranged from 14.8% to 26.4%. Judging from estimation error and the stability index of each method, the Kapur et al. method appeared to be not suitable for bruise region segmentation in this study. The other three thresholding methods were all suitable for bruise region segmentation. The moment preserving method worked best among the three methods. The developed image segmentation method not only facilitates the objective estimation of bruise volume in fruits but also effectively reduces manual efforts in segmenting bruise region from magnetic resonance images.