项目名称: 基于稀疏与自相似表征的图像插值技术研究
项目编号: No.61202150
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 计算机科学学科
项目作者: 郭强
作者单位: 山东财经大学
项目金额: 23万元
中文摘要: 图像插值是数字图像处理、遥感监测、高清视频传输和计算机视觉等领域的共性科学问题和核心技术,具有广泛的应用需求和良好的产业前景。图像边缘和纹理等细节成分的有效处理是当前未被很好解决的热点和难点问题。本项目以此为切入点,针对图像插值过程中无法有效保持图像细节成分的问题进行研究,探索具有细节保持能力的图像插值技术。重点研究图像平滑成分和细节成分分解的理论和方法、细节成分的方向分解和方向保形插值方法以及具有细节保持能力的自适应保形稀疏滤波三部分,实现一个包含这些关键技术的图像插值原型系统。通过将上述三部分研究内容融合成一个有机整体,最大限度地利用图像的稀疏性和自相似性所提供的各种信息,以此来解决插值图像的细节保持问题。本项目的研究成果将为解决图像插值技术细节保持能力差的问题提供新思路、新理论和新方法,为图像插值在各领域的有效应用提供简单实用、稳定有效的新技术。
中文关键词: 插值;低秩先验;非局部自相似性;有理函数;稀疏性
英文摘要: Image interpolation plays an important role in various application fields such as digital image processing, satellite remote sensing, high definition (HD) video transmission, computer vision and many others. It has numerous application requirements and good industry prospects. Preserving edge and texture details is a challenge to image interpolation algorithms. In order to improve the performance of preserving image details, this project explore a new detail-preserving nonlinear interpolation technique based on the sparse and self-similar properties of images. Therefore, this study mainly focus on the theory and method for image component decomposition, the shape-preserving technique for texture component directional decomposition and interpolation, and the adaptive sparse filtering for distortion reduction. We also study the relative numerical methods for the aforementioned theories and develop a prototype system that allows a user to enhance image resolution. By fusing three key parts: image component decomposition, shape-preserving interpolation and adaptive sparse filtering, we make full use of sparse and self-similar properties of images to improve the perceived visual quality of the image. The research results of this project will provide a new theory and method for detail-preserving image interpolation, a
英文关键词: Interpolation;Low-rank prior;Nonlocal self-similarity;Rational function;Sparseness