快速在线主动学习的图像自动分割算法
发布时间：2019-07-16 11:46
【摘要】：提出经前馈神经网络快速在线学习、构建像素分类模型进行图像分割的算法.首先利用谱残差法计算像素显著度,通过对少数高显著度点的分布进行多尺度分析,获得符合人眼视觉特性的显著图和注视区域.然后从注视区域和非注视区域随机抽样构成由正负样本像素组成的训练集,在线训练一个两分类的随机权前馈神经网络模型.最后使用该模型分类全图像素,实现图像分割.实验表明,文中算法在谱残差法基础上提升对图像中显著目标的分割性能,分割结果与人类视觉感知匹配度较好.
[Abstract]:A fast online learning algorithm based on feedforward neural network is proposed to construct pixel classification model for image segmentation. Firstly, the spectral residual method is used to calculate the pixel saliency. Through the multi-scale analysis of the distribution of a small number of high saliency points, the salient map and fixation region in accordance with the visual characteristics of the human eye are obtained. Then a training set composed of positive and negative sample pixels is constructed from the random sampling of the fixation region and the non-fixation region, and a two-classification stochastic weight feedforward neural network model is trained online. Finally, the model is used to classify the whole picture pixels, and the image segmentation is realized. The experimental results show that the proposed algorithm improves the segmentation performance of significant targets in the image on the basis of spectral residual method, and the segmentation results match well with human visual perception.
【作者单位】： 中国计量大学信息工程学院;
【基金】：国家自然科学基金项目(No.61572449) 浙江省自然科学基金项目(No.LY13F010004)资助~~
【分类号】：TP391.41
，
本文编号：2515058
[Abstract]:A fast online learning algorithm based on feedforward neural network is proposed to construct pixel classification model for image segmentation. Firstly, the spectral residual method is used to calculate the pixel saliency. Through the multi-scale analysis of the distribution of a small number of high saliency points, the salient map and fixation region in accordance with the visual characteristics of the human eye are obtained. Then a training set composed of positive and negative sample pixels is constructed from the random sampling of the fixation region and the non-fixation region, and a two-classification stochastic weight feedforward neural network model is trained online. Finally, the model is used to classify the whole picture pixels, and the image segmentation is realized. The experimental results show that the proposed algorithm improves the segmentation performance of significant targets in the image on the basis of spectral residual method, and the segmentation results match well with human visual perception.
【作者单位】： 中国计量大学信息工程学院;
【基金】：国家自然科学基金项目(No.61572449) 浙江省自然科学基金项目(No.LY13F010004)资助~~
【分类号】：TP391.41
，
本文编号：2515058
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2515058.html
