The Study of Effective Texture Synthesis and Image Feature Transfer

• Main Authors: Guan-da Huang, 黃冠達
• Other Authors: Chung-Ming Kuo
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/80930190998875244920

碩士 === 義守大學 === 資訊工程學系碩士班 === 97 === Recently, the technique of texture synthesis is widely used in various applications such as texture transfer, image inpainting, and image analogies. However, the synthesis quality and efficiency usually can not achieve simultaneously. Therefore, it became an important issue to be studied. In this thesis, we first focus on synthesis of perspective-distorted texture. By Hough transform, we proposed an automatic synthesis approach which provides an effective and accurate solution to synthesis line structure based perspective-distorted texture. Using Particle Swarm Optimization (PSO), the computational complexity also reduced significantly. On the other hand, the thesis also developed an effective texture transform method for artistic style transfer. Because there are many different features in an image, such as line, bright, hue, saturation, etc. therefore, we present a feature based artistic styles transfer algorithm, which employs patch-based texture synthesis approaches, as well as speeds up texture synthesis process by PSO during matching search process. We add some new feature constraints to the current algorithm, therefore, output image has similar visual effect as a manual painting. Once a certain sample image with artistic style is presented, the algorithm can accomplish image analogy by transferring the style to the target.