基于虚拟图像金字塔序列融合的快速图像增强算法  

A Fast Image Enhancement Method by Virtual Image Pyramid Sequence Fusion

作　　者：戴霞[1] 李辉[2] 杨红雨[2] 张军[3] 

机构地区：[1]西华大学数学与计算机学院,成都610039 [2]四川大学计算机学院图形图像研究所,成都610064 [3]上海交通大学航空航天学院,上海200240

出　　处：《计算机学报》2014年第3期602-610,共9页Chinese Journal of Computers

基　　金：西华大学重点实验室开放基金(szjj2011-025);江苏省自然科学青年基金(BK20130158)资助~~

摘　　要：针对图像增强算法的快速计算问题,文中提出一种"全局尝试、局部精炼"的图像增强策略,利用虚拟图像金字塔序列融合方法构建出一种具有高度可扩展性和线性时间复杂度的图像增强算法框架.在此框架下,输入图像先经过各种全局增强重映射函数的处理而形成备选图像序列,再依据像素级保边缘图像融合规则从该图像序列中精选出有效像素信息,形成最终的图像增强结果.为避免光环效应,所有图像融合计算都在输入图像的拉普拉斯金字塔分解域上进行.通过对局部拉普拉斯滤波算法进行加速的实际应用表明,文中提出算法框架可导出一种快速保边缘图像增强算法,其处理效果与原局部拉普拉斯滤波算法极为相似,但计算速度提高了两个数量级.To accelerate the implementation of the image enhancement method, we proposed a new global-to-local strategy for fast image enhancement. Based on this practical strategy, a scalable and fast image enhancement framework is proposed to construct an implementation of the existing image enhancing filter in O（N） time using the virtual image pyramid sequence fusion. Under this framework, the input image is first remapped globally by a series of tentative remapping functions to generate a virtual candidate image sequence, which is then recombined locally to a single output image under a special edge-aware pixel-level fusion rule. To avoid halo artifacts, both the output image and the virtual candidate image sequence are transformed into multi-resolution pyramid representations. Application ＇on the local Laplacian filter indicates that the proposed framework can induce a very fast approximation of this edge-preserving filter. This fast approxi- mate method is able to generate very close results to the original local Laplacian filter and achieves about two orders of magnitude speedup.

关 键 词：保边缘图像处理 图像金字塔 序列图像融合 函数逼近中圈法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]