摘要
鉴于当前算法不能很好解决重构效果和算法复杂度之间的矛盾,提出了一种基于分割的图像超分辨率重构算法。首先提出了一种基于纹理的图像分割方法,将图像分为纹理较多和较少两个区域,然后针对纹理较少区域提出了改进型小波多尺度插值方法,纹理较多区域提出了固定训练集神经网络方法。本算法综合了小波方法的简单性和神经网络方法的精确性。实验结果表明,新算法重构效果良好,复杂度较低,操作性好。
A new algorithm for super-resolution image restoration based on the segmentation is proposed to resolve contradictions of the effect and the complexity of the current algorithm. Firstly, an image segmentation method is presented to obtain two areas according to the quan- tity of the texture. Then, for the less texture area a progressive wavelet multi-scale interpolation method is used. For the more texture area a neural network method based on the invariable learning set is introduced. So the new algorithm uses the characteristics of the wavelet and the neural network. Experimental results demonstrate that the new algorithm has the properties of good vision effect, less complexity and simple operation.
出处
《数据采集与处理》
CSCD
北大核心
2005年第4期398-402,共5页
Journal of Data Acquisition and Processing
基金
江苏省自然科学基金(BK2004151)资助项目
关键词
图像分割
图像重构
超分辨率
小波
神经网络
image segmentation
image restoration
super-resolution
wavelet
neural network
