從多組旅遊照片中建構出代表性的照片集

Abstract

在本論文中，我們提出一個自動的方法有效的對照片進行管理，目標是從一起出遊的朋友們所拍攝的多組旅遊照片中建構出代表性的照片集，而這類型照片的特性是彼此間會有很高的相似度。首先，我們提出一個基於生物資訊學上比對基因序列之間相似度的演算法，並將其應用在相似照片序列的比對上。接著利用比對的結果將兩組不同的照片序列進行有效的合併。接著，我們進一步的提出一個照片評分方法，對合併後的照片序列進行評分的工作，最後保留分數較高的照片來當做代表性的照片集。有別於以往對照片評分的方法，皆是對單一照片直接進行評分的工作，我們則是利用了相似場景中其它相似的照片來輔助對單一照片的評分。在實驗上，我們收集了22個使用者所拍攝的照片來驗證我們所提出的方法。In this paper, we develop automatic approaches for effectively managing photos. In particular, we aim to find a representative stream of travel photos taken with friends in the same trip. One characteristic of such photos is that they tend to share similar contents despite the fact that they are taken by different people. Based on the observation, we first propose a method to combine multiple photo sequences into a photo stream. It is inspired by sequence alignment techniques in bioinformatics for finding similar DNA and protein sequences. Next, we develop an aesthetic quality assessment method to score each photo. Photos with higher scores are retained and formed the representative photo stream. Unlike conventional approaches that solely use information within the photo under evaluation, we show that information of the near duplicate images—which are mostly available in travel photos—could be used to facilitate the quality assessment problem. Experimental results using travel photos collected from 22 users validate the effectiveness of proposed approaches.