近年來,網路拍賣儼然成為一種新的消費模式,使用者可以在網站上搜尋欲購買的商品。但是現今的搜尋系統大多是以文字做為搜尋的依據,如果使用者無法對商品以文字確切的描述,可能會造成回傳結果不如使用者預期,所以讓使用者以影像找影像的方式是未來的趨勢。 本論文的影像搜尋系統是使用物件偵測的方式建立影像資料庫,先對影像偵測物件所在的區域,再擷取影像特徵,其可避免背景雜訊的干擾。本論文使用的兩種物件偵測的方法分別為 AdaBoost 偵測器和 Edge + PCA 演算法。AdaBoost 偵測器是特定物件的偵測演算法,可以從大量簡單特徵為主的弱分類器裡,經過 AdaBoost 演算法,合併有效的弱分類器形成一個快速並精確的強分類器。Edge + PCA 演算法是先對影像做邊緣偵測,再利用 PCA 演算法 (Principle component analysis) 找出影像中邊緣分佈密集的區域。實驗將對兩種物件偵測方法建立的影像搜尋系統進行比較。
In recent years, the development of the Internet has introduced a new means for consumers to search and buy desired products, that is, online auction and shopping websites. But today, most search engines are based only on text. If a user is unable to describe the desired products in precise keywords, the returned results will most likely not satisfy the user. Therefore, CBIR (Content based image retrieval) will be a dominating trend in the future. In this thesis, the image database is built through object detection, which then is applied in the image search system. To avoid the influence of noises from the background, we first detect the area where the object is in the image, and then the image features are extracted from object region. There are two approaches for object detection in this thesis: AdaBoost detector and Edge+PCA algorithm. AdaBoost detector is an algorithm used to detect objects of a specific class. Several weak classifiers are combined into a fast and accurate strong classifier via AdaBoost algorithm. Edge+PCA algorithm detects the edges of an image first and then finds an ellipse region where most edges appear by PCA (Principle component analysis) algorithm. To find images similar in both color and texture to the query image, we use SIFT (Scale-invariant feature transform) to re-rank the returned images. The performance of image search systems based on AdaBoost detector Edge+PCA algorithm, and re-ranking are evaluated respectively.