| Abstract: | 資訊科技提升多媒體技術進步，而數位藝術創造也隨之呈現多樣的表現方法。本論文目的在於提供一套多媒體技術，使平面藝術創作者能快速產出近年來廣為使用的圖片馬賽克(photo mosaics)作品。分析此類作品主要使用影像處理與多媒體技術，包含了影像特徵分析、影像索引及取回、色彩比對、訊號轉換、影像合成、多媒體資料庫建立等相關理論。產生圖片馬賽克技術，除了由影像資料庫檢索影像，並要考慮整張作品的視覺呈現效果。而現有文獻所使用的影像特徵並未考慮線條、色彩分佈、紋路呈現與區塊間的連續性等問題。使得搜尋而得的區塊影像相似性不高，常有突兀的感覺出現。所以，本論文將就現有方法的不足提出相對應的理論技術改進這些缺點。 首先，我們根據人類視覺對影像認知原理，利用小波轉換理論來分析影像。藉由不同頻率的影像對人類視覺呈現不同效果的原理，將影像分解成數個子頻率做為該影的特徵，而該特徵值將足以充分描述影像的色彩分佈、紋路與線條。並利用比對各不同影像間相同子頻率位置的資訊，來建立一個以小波轉換為基礎的圖片馬賽克建構技術。其次，改變以往僅以單一目標區塊為考量的比對方式，採用以整體影像效果的比對方法來取得最相似的影像，其主要考濾量到圖片馬賽克整體紋路、連續色彩及影像邊緣效果。應用的概念是以小波轉換的基底函式將相鄰的影像點做為計算上一層係數，而在經過層層的轉換之後，相鄰近的區塊會結合在一起，並呈現出整體特徵屬性。我們利用比對這些共同的特徵屬性來取得最佳的替代影像。最後，為了增加圖片馬賽克呈現的效果及去除圖片馬賽克中的格線，我們提出了使用拉普拉斯金字塔影像合成技術來將圖片馬賽克中的分割線去除。文中我們提出了能應用於圖片馬賽克的建構技術的修改方式，讓圖片馬賽克能呈現出另外一種有別於以往作品的視覺效果。 綜而言之，本論文針對圖片馬賽克問題提出全新相關技術。歸納出本論文的貢獻包含: (一)利用人類視特性，使得拚合的圖片馬賽克與原始目標影像間有較高的相似度。(二) 以整體視覺效果來考量圖片馬賽克的建構，使得圖片馬賽克在區塊間的相似性較、紋理的表現及影像邊緣的上有極佳的呈現。(三)利用影像合成的技術將圖片馬賽克中分格線去除，創作出另一種圖片馬賽克表現方式。相關的實驗結果也確實驗證了本論文所提出的理論在整體紋路、連續色彩及影像邊緣呈現上確實較目前所有文獻的方法來得佳。
The advancement of the multimedia technology is based on promoting of information technology, and the digital arts creation is also presented with various forms of expression. The purpose of this article is to provide a multimedia theory for artist to create the photo-mosaics, which is popular among art style in the last years. The relative image processing technologies of photo mosaics are included image features analyzing, image indexing and retrieving, color matching, signal transformation, and multimedia database creation. For this technique that creation of photo mosaics, not only searching image form database but also displaying visual effect has to be concerned. The selected image features of the existing approaches of photo mosaics did not consider those problems of image edges, color distribution, texture display, and continuity between blocks. And those drawbacks cause less continuity of neighboring blocks and having some suddenly perception. Therefore, the article presents an applicable theory and technique to improve those drawbacks. First, according to the image understanding of human vision system, wavelet transform is used in the approach. Due to the subband images can cause different effects of human vision, the target image is first decomposed to several subband images, which characterize into color, texture and edge features. Then, by comparing values in same position of subband between two different images, a photo mosaics technique was built based on wavelet transformation. Secondly, this article considers the important of the globally visual effects of the photo mosaics, and the effects include global texture, continuous edges, and continuous color presenting. We proposed a novel matching method that searches images by a single target block in the past, and use the idea that searches image by global effects of image to get the best similar image. The prime conception is that scaling function in wavelet transform can count next coefficient by neighboring image pixels. After several transformations, the neighboring block will combine together and present the corresponding features. Then, by comparing the corresponding features, we can get the optimum replacement image. Finally, the photo mosaics always present obvious block effect because it is composed by many block images with the same size. To remove the block effect and let the approached mosaic image likely the original image, Laplacian-pyramid was used to erase block effect and display new visual result. In conclusion, the article proposes several techniques about creation of photo mosaics. To induce our contribution in this paper includes that: 1) using the property of human vision approached mosaic image likely the original image; 2) creating photo mosaics by globally visual effects which excellently approached mosaic image; 3) using mosaic technique to remove block effect on photo mosaics and display new visual result. The experimental results have shown that our methods neither are better than the previous methods nor matter on global texture, continuous color displaying, and image edges presentation. |
