基于显著性检测的图像简化  

Image simplification based on saliency detection

作　　者：刘浩[1] 罗斌[1] 张良培[1] 

机构地区：[1]武汉大学测绘遥感信息工程国家重点实验室,湖北武汉430079

出　　处：《云南民族大学学报（自然科学版）》2016年第3期257-263,共7页Journal of Yunnan Minzu University:Natural Sciences Edition

基　　金：国家自然科学基金(61571332;61261130587)

摘　　要：图像简化作为机器视觉、计算机视觉中的一项重要任务,对于提高网络传输效率、加速视觉算法分析处理具有重要意义.针对传统图像简化模型中存在的目标、背景不加区分及尺度效应等问题,提出了一种新的基于显著性检测的图像简化模型,模型首先采用MB+显著性检测算法计算图像的目标显著区域,然后在CIE Lab颜色空间中对L波段上的背景区域进行快速水平集变换,最后按照设定准则合并背景中的非主导区域并输出简化后的图像.Image simplification is a major task in machine vision and computer vision, and has a significant impact on network transmission efficiency as well as more effective analysis and processing of vision algorithms. Because the existing problems in the current image simplification models are related to the vague distinction between the tar- get and the background region as well as scale effects, this paper proposes a novel image representation model based on saliency detection. The model first detects the salient target regions on the original image with MB ＋ sali- ency extraction algorithm, followed by the fast level sets transform of the background＇ s band in the CIE Lab color space; finally it merges non - dominant regions of the background with the pre - stated rules and regulations and outputs a simplified image.

关 键 词：显著性检测 最小障碍距离 快速水平集变换 图像简化 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]