| Summary: | 碩士 === 國立中興大學 === 電機工程學系所 === 94 === In this thesis, we present an algorithm for image inpainting using edge detection and texture analysis. Because an image sometimes has bored objects or noise, we want to remove them and get the visually plausible result. Our algorithm pays much attention on propagate the linear structure, and use inpainting algorithms filling remove regions. And then, we apply the proposed image inpainting scheme on TV logo removal. A new block-based logo detection algorithm is proposed in this work to efficiently detect various kinds of logos in any position in TV videos. The proposed system employs several observations such as logo present, shot distribution knowledge and logo characteristics to detect the real TV station logo, and exclude subtitles in the TV broadcast video. Finally, our inpainting technique is employed to inpaint the detected logo region in order to obtain the restored image. According to experimental results, our proposed system of the TV logo detection have been tested on several hours of real TV videos and achieving 99% correct labeling.
|
|---|
