电器标签分类的SVM方法研究

Appliance Label Classification Based on SVM

SVM是图像识别与分类中的重要方法。对家电标签图像的自动分类问题进行研究,设计了图像采集、预处理、特征提取、特征向量分类整套算法流程。对除噪后的标签图片,利用标签矩形边框这一信息校正因标签拍摄角度而引起的畸变。选取校正后标签图像的HSV统计直方图、ASM能量、逆差矩、对比度和自相关性5项参数构成特征向量对图片进行描述。采用决策树+SVM分类器的结构对特征向量进行分类,最终获取标签图像所属类别。实验结果表明,决策树+SVM结构在训练样本个数极少的条件下,仍能完成模型训练,并以一定的准确率快速完成目标图像的分类工作。

SVM is an important method in image recognition and classification. The automatic classification of household appliance label image was studied, and the flow of image acquisition, preprocessing, feature extraction and classification was designed. The perspective information was obtained using the rectangular border of the label so that the distortion caused by the angle of the label shooting can be corrected. The HSV statistical histogram, ASM energy, inverse moment, contrast as well as autocorrelation were selected to construct the image feature vector. The decision tree +SVM classification structure was used to classify the feature vectors. Show by experimental results, the decision tree + SVM structure can still classify the target image quickly and correctly with a certain accuracy under the condition of a small number of training samples.