賴宗煊

Tsung-HsuanLai

碩士論文（2012）

支援向量回歸為基之具預測能力影像補全技術研發

Image Completion Using Prediction Concept Based on Support Vector Regression

關鍵字 Keywords

影像補全、影像修復、紋理合成、結構預測

image completion、image inpainting、texture synthesis、structure prediction

摘要

影像補全(Image completion, IC) 是將受損影像修復或移除物件的技術，已廣泛地應用於生活當中。本研究透過導入預測之概念，不僅能夠透過系統自動選取出影像中之受損區域，並藉由結合影像補全與預測之概念，提出一利用支援向量回歸(Support vector regression, SVR)進行結構預測之影像補全技術。透過改善過去方法，以及加上預測結構的方式，並在最後進行失真樣本的偵測及改善，以有效地提升修復影像品質。實驗結果亦證明了能夠有效修復大多數之受損影像，且能夠為人眼所接受。

Abstract

Image completion is a technique widely used that automatically removes objects or repairs damaged portions of image. However, information regarding the original image is often lacking in structure reconstruction, and as a result, images with complex structures are difficult to restore. This study proposed a SVR-oriented Image Completion (SVR-IC) method, the goal of which is to predict the original structure of unknown areas and then repair or make appropriate adjustments to the structure and texture of the damaged area. From the experimental results, SVR-IC produced images of good quality that were superior to those of other methods. The results show that integrated structure prediction to image completion can effectively enhance the quality of the restored image.