多任务学习框架下的单帧图像超分辨率重建
吴亚榕
【期刊名称】《仲恺农业工程学院学报》 【年(卷),期】2015(000)004
【摘要】For making full use of the property of self similarity, a novel SISR method based on a variant of Gaussian process regression ( GPR) was proposed by considering it as a multi-task learning problem. This method could efficiently suppress the artifacts along the salient edges and produce natural looking ap-pearance in the final high resolution ( HR) results.Extensive experimental results demonstrated that the results produced by our method could be equivalent or superior to other state of the art algorithms.%为了充分利用图像的自相似性质，针对图像的超分辨率重建问题构造了一个多任务学习的问题，并基于高斯过程回归进一步扩展，提出了一个新颖的单帧图像超分辨率重建算法。该算法能够在图像的超分辨率重建结果中有效地抑制图像显著边缘处出现的噪声和伪影，并能生成视觉效果更为自然的高分辨率图像。实验表明该算法产生的高分辨率图像结果能够相当于甚至超越当前先进的算法。 【总页数】4页(50-53)
【关键词】单帧图像超分辨率;多任务学习;高斯过程回归;自相似 【作者】吴亚榕
【作者单位】仲恺农业工程学院科技处，广东广州510225 【正文语种】中文
多任务学习框架下的单帧图像超分辨率重建
