基于线条识别的图像语义分类方法  

Image semantic classification method based on line identification

作　　者：李冰季[1] 金聪[1] 

机构地区：[1]华中师范大学计算机学院,武汉430079

出　　处：《电子测量技术》2013年第3期53-57,共5页Electronic Measurement Technology

基　　金：教育部人文社会科学研究规划基金(11YJAZH040);武汉市科技攻关计划(201210121023)的资助项目

摘　　要：图像的底层特征与高层语义之间存在着"语义鸿沟",机器不能由图像点阵或图像底层特征而推导出图像的语义。大多数算法中,图像底层特征的提取往往只关注图像单个点或者局部的特性。从基于点与点相互关系的角度来处理图像的语义分类,先简要介绍当前图像语义领域的相关研究方法,之后根据模式组合的思想,提出识别图像中轮廓线条的方法。算法首先利用Canny算法将图像转换为线条轮廓,然后提出一种从图像中识别不同线条的方法,并根据线条线长和角度变化率的分布范围来对不同类别的图像语义进行分类。实验证实了算法的有效性,且在仅有少量样本的情况下仍可获得较高的识别率。There exists a semantic gap between the low-level visual features and the high-level image semantic, and machine can＇t directly understand the points or the low-level visual features generated by computer. Current methods of extracting visual features tend to focus on isolated points and local characteristics. In this paper,we propose a method, based on the relationship between points and points, to extract visual features of images. This paper first briefly introduces some research methods in the image semantic field, and then proposes the recognition image contour line method based on mode combination. We use Canny algorithm to convert image into line profile, and then propose a method to identify different lines in image, finally according to the distribution range of line length and angle rate, we classify images to different types. Through an experiment, we confirm the effectiveness of this method. This experiment gets a high accuracy ratio only needing a small amount of samples.

关 键 词：图像语义 语义分类 线条识别 CANNY算法 角度变化率 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术]