摘要
针对动物图像的分类识别问题,提出了一种基于Adaboost分类器的动物二分类识别方法。首先对样本图片进行边缘特征提取,选取八种具有显著形状不变性的特征描绘子,并对其合理性和优越性进行验证。后用Adaboost分类器对所得特征矩阵进行训练,得到最有效的分类特征,并对从Shape Database形状图片库中选取三组动物图像进行十折交叉验证实验。狗和牛、牛和象、青蛙和牛的正确分类识别率分别达到85%、90%和92.5%。实验表明该分类识别方法能较准确进行二分类识别,是一种较有效的动物图像二分类识别方法。
To solve the classified problems of animal＇s images,a classified method based on Adaboost is designed for dichotomic recognition.First,edge features of sample images are extracted.Then,eight characteristic descriptors having significant shape invariances are selected and their rationalities and superiorities are tested.Adaboost classifier to train the matrix of characteristics,aiming to get the most effective classifying feature.Experiment on the three groups of animals＇images selected from the photo gallery called Shape Database through 10-fold cross-validation.The identified rates of classification of dogs and cattles,cattles and elephants,frogs and cattles reach 85 percent,90percent and 92.5percent respectively.The experiment shows the classified method can classify images comparatively accurate into two sorts and it is rather definitely an effective way to classify animals＇images into two categories.
出处
《计算机与数字工程》
2017年第4期720-726,767,共8页
Computer & Digital Engineering
基金
国家级大学生创新创业训练计划资助项目(编号:201610145014)
中国博士后科学基金资助项目(编号:2016M591446)
中央高校基本科研业务费(编号:N140503004)
国家自然科学基金青年科学基金资助项目(编号:61402097)资助
