使用型態篩選法及動態部分函數作影像擷取

Abstract

近年來，多媒體資料的大量成長。使得如何發展對這些資料作儲存、瀏覽、檢索以及擷取的技術成為一項重要的議題。在影像擷取的研究已有許多，但鮮少有使用到因為能即時應用而著稱的數學型態學。因此，本篇論文的目的是改善由J.S.Wu提出的morphological primitives的準確度，並且提出另一個型態運算子—篩選法作為影像的特徵。實驗結果顯示，morphological primitives使用動態部分函數為相似度量測最佳有30%的準確度改善。而另一方面在大部分的影像中，granulometric histogram比起color histogram，color moment primitives，以及morphological primitives有較佳的準確度。

In recent years, there is an explosive growth of multimedia data. How to develop techniques for storage, browsing, indexing, and retrieval them efficiently is a significant issue. There are many researches on image retrievals, but very few usage of mathematical morphology which is impressive because of making real-time applications possible. Our goal is to improve precisions of morphological primitives which was introduced by J.S.Wu and propose another morphology operator: morphological granulometry as an image feature. From experimenal results, image retrieval using morphological primitives with dynamic partical function have improvement up to 30%. On the other hand, granulometric histogram has better precisions than those of color histogram, color moment primitives, and morphological primitives in most cases.