樹狀物體之分支分析與萃取在文心蘭分級之應用

Abstract

一般而言花束總是一批一批地被呈獻給消費者，它的外觀對消費者的購買決定具有舉足輕重的影響。因此我們考慮應用圖形識別分級文心蘭的可能性，利用影像切割文心蘭原圖得到它的外型，其樹枝部份的影像類似樹狀物體。我們應用Branch-based region growing演算法，指出樹狀物體的頂點。其中為了解決相似接觸的分支還有交叉分支的情形，我們提出morphological thinning影像切割與圖形學分支分析的理論。並且經過一些簡化與計算的過程達到最後估測主幹長度、以及各個分支長度的實驗結果。

Posies are normally presented to consumers in batches. The appearances of these have powerful effect on consumer decision. For this reason, oncidium grading system using pattern recognition is considered. The oncidium shape is segmented from original image. It’s a tree-like image. The “Branch-based region growing” algorithm is applied. It figures out vertices of tree-like image. In this paper, the morphological thinning and branch analysis are presented for solving ambiguous touch and case of crossed branches. After some simplifications and calculation, feature parameters such as length of trunk, length of each branch, count of branch etc. are extracted. Experimental results from artificial and real image are included to demonstrate the validity of the proposed grading methods.