| 摘要: | 直方統計圖的比對是圖形識別的相關應用中常需使用到的技術，比對兩個不同的圖樣可藉由比對相對應的直方統計圖來完成。在本篇論文中，我們將深入分析直方統計圖的特質，並且將之應用在影像檢索技術的提升。我們所探討的主題依次是 一、直方統計圖的空間包容度的分析。在此我們提出了一個公平的比較方法，可以利用直方統計圖的空間包容度的量測來比較各種不同特徵的直方統計圖或不同量測函數的鑑別能力。 二、直方統計圖的模糊化技術。藉由模糊化技術可以克服由於直方統計圖統計圖的鑑別力提升卻造成相似圖樣的距離同時被拉大的副作用。 三、自動化的相似度判斷機制。可以取代傳統上以人工方式評估影像檢索系統效能的缺點，是一個客觀而且能夠量化統計分析影像檢索檢索效能的技術。 四、大型直方統計圖的比對演算法。我們提出了一個適用於大部分常見量測函數與直方統計圖的大小無關的比對演算法。藉由我們的技術，直方統計圖的鑑別能力可藉由提高特徵數或各個特徵的解析度自由提升，而不必擔心直方統計圖大小對效率的影響。 最後本論文以數個影像檢索的實驗分析驗證我們提出的方法，並做出總結。 Histogram matching is a commonly adopted technique in the applications of pattern recognition. The matching of two patterns can thus be accomplished by matching their corresponding histograms. In this dissertation, we comprehensively analyze the characteristics of histogram and discuss the issues frequently appeared in image retrieval by utilizing the detail analyzing results. The characteristics inherent in histograms that we study are histogram capacity and histogram smoothing. Through the analysis of histogram capacities, researchers can evaluate and compare the discrimination capabilities of different histogram measurements or histograms embedded with different features. As to the histogram smoothing technique, it can compensate the penalty resulting from the enlargement of histogram dissimilarities of similar patterns due to the increase of discrimination capability. Moreover, an automatic similarity judgment mechanism without human involving for the evaluation of the retrieval effectiveness of image retrieval algorithms is proposed. Traditionally, researchers usually use the statistic scores of similarity judgments deriving from a few persons to demonstrate the performance of their image retrieval systems. However, the statistic scores of similarity judgment are too subjective. Through the use of our proposed similarity judgment mechanism, the retrieval effectiveness of image retrieval systems can be evaluated objectively, transparently and quantitatively. As we know, the number of features and the resolution of each feature will determine the size of histograms. Usually, the more the number of features and the higher the resolution of each feature, the stronger the discrimination capability of histograms will be. Unfortunately, it will lead to the decrease of the efficiency of histogram matching because traditional algorithms in evaluating similarity are all relevant to the histogram size. In this dissertation, a novel histogram-matching algorithm is proposed whose efficiency is irrelevant to the histogram size. By adopting our proposed algorithm, future researchers can focus more on the selection and combination of histogram features and freely adjust the resolution of each feature without worrying the decrease of retrieval efficiency. To demonstrate the feasibility and validity of the analyzing results and the proposed algorithms, a similar image retrieval experiment, which was constructed based on multi-feature histogram, is conducted. Although the experiment itself is the image retrieval application, the proposed histogram-relating techniques are not merely limited to be employed to image patterns. They can be applied to other retrieval-like applications which are independent of images or even other fields irrelevant to retrieval-like applications. |
