基於內容影像檢索技術的基礎是影像特徵的提取。一般來說,影像特徵的表示均是高維向量,對於大型的影像資料庫,高維向量的存儲以及高維空間中距離的計算,其空間複雜度和運算複雜度非常高。 用於影像資料庫的CBIR系統中使用的技術主要集中在形狀提取。因此,使用形狀描述符zernike矩來表徵影像檢索系統中的影像是明智的。本研究的目的調查現有降維技術中應用在影像檢索系統的最佳技術。本研究創建了一個具有1,000張圖片的影像資料庫。本研究中使用的PR曲線(精確率和回收率曲線)評估系統檢索效能。實驗結果顯示,高斯核心主成分分析(Gaussian Kenel PCA)降維技術的影像檢索效能最佳。在未來的工作中,將使用其他降維技術來比較其有效性。
Based on the content image retrieval technology is based on the extraction of image features. In general, the representation of image features is a high-dimensional vector. For large-scale image library, high-dimensional vector storage and distance calculation in high-dimensional space, the spatial complexity and computational complexity are very high. The techniques used in CBIR systems for image databases are focused on shape extraction. Therefore, it is wise to use the shape descriptor zernike moments to characterize the image in the image retrieval system. The purpose of this study is to investigate the best techniques used in image retrieval systems for existing dimensionality reduction techniques. This study creates an image library with 1,000 images. The PR curve (precision and recall curve) used in this study assess system performance. The experimental results show that the Gaussian kernel principal component analysis(Gaussian Kenel PCA) has the best image retrieval performance. In future work, other dimensionality reduction techniques will be used to compare their effectiveness.