基于兴趣点密度加权的图像美学质量评价

An image aesthetic quality evaluation based on density of interest points weighting

针对现有方法没有充分考虑图像空间美学信息,并且评价效果过分依赖于主体区域识别效果的问题,提出一种基于兴趣点密度加权的图像美学质量评价模型。首先对图像进行超像素分割和兴趣点检测并提取特征描述子,然后统计超像素块内的兴趣点个数,根据兴趣点密度对超像素块内的特征描述子进行加权并进行局部约束线性编码处理,最后利用机器学习方法进行图像美学质量评价。实验结果表明该方法用来图像美学质量评价大大减少了特征维度和计算时间,提高了评价模型的准确率。

The aesthetic evaluation model has low accuracy since the existing methods heavily rely on the performance of object region extraction and insufficient characterization of image space aesthetic information.Aiming at solving the problem,an image aesthetic quality evaluation model based on density of interest points weighting is proposed.First,su.perpixel segmentation was used to segment images into irregular superpixel blocks,and interest points were detected and feature descriptors were extracted on the original image;Second,the density of interest points in the superpixel block were calculated,and the feature descriptors were weighted according to the density of interest points and coded by localityconstrained linear coding method.Finally,the machine learning method was used to evaluate the aesthetic quality of im.ages.The experimental results show that this proposed method is effective for image aesthetic quality evaluation,which greatly reduces the feature dimension and the computation time,and improves the accuracy of the evaluation model.