自然场景兴趣区的分量组合-压缩快速分割法  

Fast segmenting ROI of field scenario based on RGB combination and bit-masked reduction

作　　者：周平[1] 钟取发[2] 汪亚明[2] 赵匀[1] 

机构地区：[1]浙江大学生物系统工程与食品科学学院,浙江杭州310027 [2]浙江理工大学视觉检测实验室,浙江杭州310018

出　　处：《浙江大学学报（工学版）》2007年第4期655-661,共7页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(50545027);浙江省高校中青年学术带头人资助项目

摘　　要：为实现野外自然道路、农田复杂场景或动态纹理感兴趣区(ROI)的快速分割,提出了一种基于RGB颜色分量线性组合与颜色位屏蔽压缩的主动摄像机视觉目标分割新技术.结合含误差带FloodFill算法与提出的RGB色域特征循环学习,获得理想的自然道路场景核心或指定纹理目标.以该ROI为对比模板,快速搜索出进行特定目标实时分割的最优RGB颜色分量组合.大量实验显示:颜色分量线性组合能增强特定目标的显示特性,颜色位屏蔽能压缩色域,提取近似的颜色主成分,同时减弱图像中的噪声影响,并减小干扰引起的纹理分割误差.该方法的计算时间几乎不受目标纹理复杂度影响,能在30 ms内全图分割分辨率为320×240的图像序列中的多目标,对颜色噪声变化和光线波动有较强鲁棒性,在野外自然道路或农田场景、多颜色目标和动态纹理兴趣区的分割速度与稳定性上取得明显进步.A novel real-time technique for segmenting image into regions of interest （ROI） of field scenario as rural road or dynamic textures in active vision was presented, which is based on linear combination of red,.green and blue （RGB） components and bit-masked reduction. The extended FloodFill algorithm combined with a ring-style learning pattern of RGB color features archived the ideal kernel regions in field scenario or interested textural objects. Compared with such a ROI template, the optimized combination coefficient of RGB components was searched out for real-time segmentation of special objects. Lots of experimental results showed that the proposed linear combination of color components successfully enhances the visual pattern, and that the bit-masked color reduction extracts out the approximate principal color components by cutting down noises in image and decreasing errors in its texture segmentation. The achieved resuits are influenced little by textural structure of regions, with full segmentation of multi-objects done within 30 ms for color image sequence of 320×240 resolutions. This method is robust against the variation of color noisy distribution and illumination, and significant progress is made in stability and it speeds up segmentation of the ROI of field scenario, colored multi-objects or dynamic textures.

关 键 词：自然场景 模板学习 RGB分量组合 色域压缩 快速分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]