多方向照片刮痕偵測技術

Abstract

近年來，由於對於老舊照片、影片的保存、數位典藏、研究用途、或是再製作的需求提升，解決將這些傳統媒材數位化面臨到的問題便日漸重要。刮痕是其中一種常見的缺陷，過往的研究方向主要是偵測老舊電影上的刮痕，由於電影撥放的特性，其刮痕主要是為水平或垂直的。但在傳統底片攝影中，刮痕就不限於這兩個方向，更還有非直線的情形發生，使此問題變得較為困難。 本篇論文提出一全自動的刮痕偵測方法。其利用刮痕的特性，先將圖片中較為銳利的部分取出，並使用材質分類 (Texture Classification) 技術將疑似刮痕的像素分別出來。最後，利用霍夫轉換 (Hough Transform) 來偵測其中直線的大致方向及位置，再利用得到的資訊取得精細的刮痕。 實驗結果顯示本論文之方法能準確地找出畫面中任何方向的括痕，對於非純粹直線的刮痕也依然能夠偵測出來。

In recent years, for the purposes that preserving, researching, re-producing old photos and films, the demand of digitizing the images which are on the traditional media is increasing. Therefore, how to deal with the problems in the digitizing process is become more and more important, and scratch is one of them. In the past, scratch detection researches try to detect the scratch mainly on the movie. Due to the properties of movie, the scratch on the movie film are mostly vertical or horizontal. But in the field of traditional photography, the scratch might in any direction, moreover, it would not be a straight line. The thesis presents a automatic method which is based on the properties of scratch. It extracts the sharp features from the damaged image, and uses texture classification technique to get the scratch part. Then Hough transform is involved to detect the rough direction of scratch. Following a refinement scheme is applied to get a precise binary scratch map. The experimental results show that our method can detect scratches in any directions, moreover, some scratches which are not straight lines also could be detected.