基于Zernike矩变换与支持向量机的不良视频检测方法  

Research on Adult Image Detection Using Zernike Transform and Support Vector Machine

作　　者：刘春[1,2] 谢长生[1,2] 朱光喜[1,2] 吴伦[1,2] 

机构地区：[1]华中科技大学计算机学院,武汉430074 [2]华中科技大学武汉国家光电实验室,武汉430074

出　　处：《武汉理工大学学报》2011年第7期133-137,共5页Journal of Wuhan University of Technology

基　　金：国家973重点基础研究发展计划(2011CB302303);国家863高技术研究发展计划(2009AA01A402)

摘　　要：为了构建基于机顶盒等嵌入式平台的不良视频识别与过滤技术,从图像的高层语义特征着手,利用Zernike矩变换和多尺度图层进行快速的形状特征粗匹配,对敏感图像关键部位进行初步识别,再采用支持向量机对形状区间的像素分布特征做进一步的判别。实验表明,结合多种快速算法,系统的处理速度满足实时要求,与传统的肤色比例检测方法相比,在提高正检率的同时能降低误判率。For constructing pornographic image detection and filter technology based embedded platform like set top box,this paper presents a combined method based high-level semantic characteristic and aims to identify specific shape feature which integrated Zernike moment transform and multi-level template match,further the pixel distribution features of possible shape areas are detected by support vector machine.The test results show that when combined fast computing algorithm,the system performance can meet the real time requirement and the correct recognition is higher than traditional method based on skin area detection,at the same time the error recognition rate is decreased.

关 键 词：ZERNIKE矩 支持向量机 图像检测 模板匹配 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]