王艳 周小平 王睿 孙冰雪



摘要:目的 研究基于视觉词袋的图像检索方法并应用于长白山中草药植物图像检索领域。方法 采用SURF算法提取图像视觉特征,稀疏编码方法构造视觉词典,并提出支持向量机(SVM)和近似最近邻(ANN)相结合的改进方法完成分类器分类训练。结果 选取2500张中草药图像作为检索样本,在视觉单词数量为500的情况下,平均检索时间为481 ms,平均查准率为88.95%。结论 本方法能有效提高图像检索效率与准确度,同时表现出较好的鲁棒性。
关键词:中草药;图像检索;视觉词袋
DOI:10.3969/j.issn.1005-5304.2018.02.021
中图分类号:R2-05;R282.74 文献标识码:A 文章编号:1005-5304(2018)02-0095-04
Research on Image Retrieval Method of Wild Chinese Herbal Medicine Plants
in Changbai Mountain
WANG Yan, ZHOU Xiao-ping, WANG Rui, SUN Bing-xue
College of Pharmacy, Jilin University, Changchun 130012, China
Abstract: Objective To study the visual word bag based image retrieval method and apply it in the field of image retrieval of wild Chinese herbal medicine plants in Changbai Mountain. Methods SURF operator was used to extract visual features. Then sparse coding method was used to structure visual dictionary. The classifier was trained by combination of support vector machine (SVM) and approximate nearest neighbors (ANN) method. Results Totally 2500 photos of Chinese herbal medicine plants were chosen. When the visual word number was 500, the average retrieval time was 481 ms, and the average query accuracy was 88.95%. Conclusion The method can effectively improve the efficiency and accuracy of image retrieval, and has better robust.
Keywords: Chinese herbal medicine; image retrieval; visual word bag
長白山位于吉林省东南部,属国家级自然保护区,森林资源丰富、植物种类繁多,共有野生植物2700余种,其中药用植物1000余种,名贵药用植物有人参、党参、东北刺参、黄芪、天麻、红景天和灵芝等,是我国著名的北药基地[1]。为高效检索和鉴别中草药植物图像,本研究探索基于内容的图像检索方法在中草药图像检索领域的应用。
传统图像检索过程中,先通过人工对图像文字标注,再用关键字检索图像,这种依据图像描述的字符匹配程度提供检索结果的方法,简称为“字找图”,既耗时又主观多义。基于内容的图像检索可克服“字找图”方式的不足,直接从待检索的图像视觉特征出发,在图像库中找出与之相似的图像,这种依据视觉相似程度给出图像检索结果的方法,简称为“图找图”[2-3]。在基于内容的图像检索中,视觉词袋模型(bag of visual word,BoVW)已成为较常见的方法。……