摘要
本文提出了污损图像的自动标注算法.首先确定待标注图像的污损区域,根据污损区域的位置和比例划分字典中的图像,提取图像的底层特征,基于底层视觉特征构建稀疏模型,确立污损图像与字典中图像的相似关系,对字典中的相似图像进行分块处理,由污损图像与字典图像的子块特征确定其可能相关的标注词.最后通过概率统计完成污损图像的自动标注.实验表明该方法在一定程度上弱化污损区域对图像标注的不利影响,较好地实现了污损图像的自动标注.
We put forward the method based on sparse expression of stained image retrieval and automatic labeling algorithm.First the image to the labeling of stained area is determined,at the same time,according to the position and proportion of corresponding classified dictionary of each image,the sparse model is set up by learning mechanism than the stained image and image similarity in the dictionary,take out the most similar 30 images and then treated respectively test 30 image block processing,extracting the characteristics of each sub-block and corresponding sub-block and image similarity coefficient under test.For each part of the stained image automaticaly tagging of work is done.experiments show that the method is largely limit integration the visual characteristics of the image and text representation,weaken the negative impact of the stained area of image annotation,better realize the automatic tagging tainted the image.
出处
《山东师范大学学报(自然科学版)》
CAS
2017年第1期30-35,共6页
Journal of Shandong Normal University(Natural Science)
基金
教育部博士点基金资助项目(20113704110001)
关键词
污损图像
稀疏模型
分块
自动标注
stained image
sparse model
defaced labeling
tagging