一种基于多特征组合和SVM相关反馈的皮肤病图像检索算法
|
摘要
针对特征复杂的皮肤病受损区域图像难以用单个特征准确表达,且低层视觉特征与高层语义空间之间存在语义鸿沟,造成皮肤病受损区域图像检索困难的问题,提出了一种基于多特征组合和SVM相关反馈的皮肤病图像检索方法。首先对预处理之后的皮肤病受损区域的图像进行多特征提取并进行组合,然后采用欧式距离相似度模型对皮肤病受损区域图像初步检索,最后引入了带有衰减系数的SVM相关反馈算法,提高皮肤病受损区域图像的检索准确率。实验结果表明,引入带有衰减系数SVM相关反馈的方法可以检索到更多的相关图像,明显提高了检索的查准率。
Aiming at the problem that the image of damaged area with complex features can not be accurately expressed by a single feature, and there is a semantic gap between low-level visual features and high-level semantic space, which makes it difficult to retrieve the image of damaged area of skin diseases, a skin disease image retrieval method based on multi-feature combination and SVM relevance feedback is proposed. Firstly, the multi-feature extraction and combination of the images of the skin lesions after preprocessing are carried out. And then the Euclidean distance similarity model was used to search the images of the damaged areas of the skin disease. Finally, an SVM relevance feedback algorithm with attenuation coefficient is introduced to improve the retrieval accuracy of images in damaged areas of skin diseases. The experimental results show that the introduction of the SVM relevance feedback method can retrieve more related images,it is concluded that which signi ficantly improves the precision of the search.
Aiming at the problem that the image of damaged area with complex features can not be accurately expressed by a single feature, and there is a semantic gap between low-level visual features and high-level semantic space, which makes it difficult to retrieve the image of damaged area of skin diseases, a skin disease image retrieval method based on multi-feature combination and SVM relevance feedback is proposed. Firstly, the multi-feature extraction and combination of the images of the skin lesions after preprocessing are carried out. And then the Euclidean distance similarity model was used to search the images of the damaged areas of the skin disease. Finally, an SVM relevance feedback algorithm with attenuation coefficient is introduced to improve the retrieval accuracy of images in damaged areas of skin diseases. The experimental results show that the introduction of the SVM relevance feedback method can retrieve more related images,it is concluded that which signi ficantly improves the precision of the search.
引文
[1]蒲晓蓉,王之骢,宋帅领.基于朴素贝叶斯分类器的皮肤病图像颜色特征提取方法[P].四川:CN106557771A,2017-04-05.
[2]王兴旺,杨慧兰.人工智能实现专业级皮肤癌诊断:未来医学发展动向[J].实用皮肤病学杂志, 2017(3):141-141.
[3]于凡,万艳丽,胡红濮.医学图像检索技术发展现状[J].中华医学图书情报杂志, 2017, 26(7):31-35.
[4]孙银辉.色素性皮肤病图像预处理与内容检索研究[D]. 2016.
[5]宋帅领.色素性皮肤病图像的特征提取与识别[D]. 2016.
[6]白婧文,赵志诚.一种新的基于SVM相关反馈的图像检索算法[J].软件导刊, 2010, 09(10):49-51.
[7]赵理君,唐家奎,于新菊,等.综合视觉特征度量与SVM的遥感图像检索方法[J].中国科学院大学学报, 2013, 30(3):347-352.
[8]顾晓东,杨诚.新的颜色相似度衡量方法在图像检索中的应用[J].仪器仪表学报, 2014, 35(10):2286-2292.
[9]闫允一,姜帅,郭宝龙.结合稳定兴趣点和Gabor小波的图像检索[J].西安电子科技大学学报(自然科学版), 2014, 41(5):118-123.
[10]刘胜蓝,冯林,孙木鑫,等.分组排序多特征融合的图像检索方法[J].计算机研究与发展, 2017(5).
[11]孙树亮,林雪云.基于记忆的SVM相关反馈算法[J].计算机科学, 2011, 38(10):256-258.
[12]朱红斌.综合颜色和纹理及SVM相关反馈的图像检索[J].计算机工程与应用, 2009, 45(5):183-185.
[2]王兴旺,杨慧兰.人工智能实现专业级皮肤癌诊断:未来医学发展动向[J].实用皮肤病学杂志, 2017(3):141-141.
[3]于凡,万艳丽,胡红濮.医学图像检索技术发展现状[J].中华医学图书情报杂志, 2017, 26(7):31-35.
[4]孙银辉.色素性皮肤病图像预处理与内容检索研究[D]. 2016.
[5]宋帅领.色素性皮肤病图像的特征提取与识别[D]. 2016.
[6]白婧文,赵志诚.一种新的基于SVM相关反馈的图像检索算法[J].软件导刊, 2010, 09(10):49-51.
[7]赵理君,唐家奎,于新菊,等.综合视觉特征度量与SVM的遥感图像检索方法[J].中国科学院大学学报, 2013, 30(3):347-352.
[8]顾晓东,杨诚.新的颜色相似度衡量方法在图像检索中的应用[J].仪器仪表学报, 2014, 35(10):2286-2292.
[9]闫允一,姜帅,郭宝龙.结合稳定兴趣点和Gabor小波的图像检索[J].西安电子科技大学学报(自然科学版), 2014, 41(5):118-123.
[10]刘胜蓝,冯林,孙木鑫,等.分组排序多特征融合的图像检索方法[J].计算机研究与发展, 2017(5).
[11]孙树亮,林雪云.基于记忆的SVM相关反馈算法[J].计算机科学, 2011, 38(10):256-258.
[12]朱红斌.综合颜色和纹理及SVM相关反馈的图像检索[J].计算机工程与应用, 2009, 45(5):183-185.