改进的Beta过程因子分析图像修复算法

An improved imageinpainting algorithm based on Beta process factor analysis

图像修复是利用图像已知信息对图像破损区域进行填充修复的过程,而非参贝叶斯技术在图像稀疏表示中被认为是一种有效的字典学习方法,作为一种有效的非参贝叶斯算法,基于Beta过程因子分析算法(BPFA)在去噪、修复以及压缩感知方面有很广泛的应用.然而现有的BPFA算法在对含噪的破损图像修复时收敛速度慢,针对这个问题本文在BFPA算法更新字典时与K-SVD算法相结合,提出一种基改进的BPFA学习算法,改进算法利用K-SVD算法简单收敛速度快的特点,在原有算法更新参数时,利用OMP稀疏编码更新字典候选集以达到提高算法的收敛速度的效果.得到的结果表明本文算法能够更好地修复含噪破损图像获得较好的视觉效果.

Imageinpainting is the process to fill and repair the damaged area of images based on the given information of the images, and Nonparametric Bayesian technique is known as an effective dictionary learning method in image sparse representation. As an effective method of Nonparametrie Bayesian, Beta process factor analysis （BPFA） is widely used in the aspects of denoising, repair and compressed sensing. However, the convergence rate of existing BPFA is slow when re- pairing the damaged images with noise. In order to increase the convergence rate, an enhanced BPFA learning algorithm is proposed, the K-SVD algorithm is combined with BPFA algorithm when updating the dictionary, which means that when updating parameters, OMP algorithm is used to update dictionary candidate set to increase the convergence rate because K-SVD algorithm is simple and has fast convergence rate. The results show that the enhanced algorithm can repair the images better and obtain a preferable visual effect.