为了分离出图像中具有不同特征的成分,结合变分与字典学习方法,提出一种图像分解模型和结构-纹理字典学习算法.首先在模型中引入字典约束项,使得结构-纹理学习字典互不相关,增强了2个字典的独立性；然后使用投影梯度下降算法给出一种带有字典约束的交替字典学习算法.实验结果表明,采用该算法学习得到的自适应字典可以有效地刻画图像的不同成分,不仅很好地分开了图像的结构和纹理,并且能去除噪声,最终得到高质量的图像分解结果.%In order to separate different features of image,this paper presents a variational model for image decomposition.Meanwhile a new cartoon-texture dictionary learning algorithm is proposed.In the model,we introduce an incoherence promoting term,which encourages different components to be as independent as possible.Using decreasing gradient optimization algorithm,an alternate dictionary learning algorithm with constraint is presented.Numerical experiments show that the learned dictionaries by the proposed algorithm can describe the different components of image effectively,and lead to high quality image decomposition and denoising performance.
展开▼
