近來數位相機普及，使得個人相片數量大增，要尋找相片時，作業系統內建之文字搜尋功能並不符合搜尋相片使用，使用者只能在資料夾之間瀏覽相片，造成搜尋效率低落。而現有之圖像檢索技術大多應用於網路上之圖像搜尋，但個人相片搜尋與網路圖像搜尋目的不同，個人相片搜尋著重尋找印象中之相片，故改進目前圖像檢索技術並進行深入探討以應用於個人相片搜尋。 本研究以基於內容之圖像檢索技術作為改良對象，系統對相片集做圖像前處理，使用者可選擇繪圖或相片給予系統提示，系統以權重調整後之特徵進行前饋式以圖找圖，幫助使用者搜尋相片。 透過整理國內外文獻，經由圖像記憶與認知、視覺傳達、色彩學與形狀學等領域，歸納出人類記憶之圖像特色為彩度明度偏高且色彩單純形狀簡潔，故將目標圖像提升彩度、亮度、減化色彩做為圖像前處理之依據，以求與印象中之圖像相符，由實驗得出能有效的幫助使用者縮減搜尋範圍，縮減效果可達8%；特徵權重調整參考人類觀察圖像時所注意之敏銳度，其中明度最強，其次為色相，之後為彩度，作為調整特徵值權重之依據，實驗得出以最佳調整權重HSV(0.75, 0.75, 1.5) 進行前饋式搜尋時，可幫助使用者以較少次數搜尋到目標，最佳表現可減少13次搜尋次數。 Due to digital camera has been generally used recently, the quantities of individual digital photos rise fast. When a photo needs to be found, the finder in OS by text is inefficient and not fit in searching photos. Most image retrieval systems are used in web image searching, but the purposes of searching individual photos are different from web images. Search in individual photos focuses on the photos which are remembered, so the image retrieval methods are improved to fit in individual photo search system. The derived system is based on Content-based Image Retrieval. In the beginning system will preprocess photo collections, and users give tips by painting or by giving a sample photo. Then the target will be found by Sequential Forward Selection according to the adjusted color feature. Preprocess photo collections mean process photos according to the image feature in human memory, the saturation and brightness is clear and the color is simply, so the saturation and the brightness shall be enhanced, and the color shall be simplified accordingly. The searching range can be reduce 8% as shown in experiments. Adjust weight of color feature means to adjust the weight of color feature according to the attention when human watches a photo, its sensitivities are Value, Hue and Saturation respectively. The best HSV weight is Hue: 0.75, Saturation: 0.75 and Value: 1.5 in experiments, it could reduce the search 13 times.
