局部模糊检测优化算法

目前基于特征的局部模糊检测算法为了优化特征响应需要在多尺度下重复计算局部模糊特征,且邻接关系复杂,导致计算量大,时间效率低.针对上述时间问题,本文提出一种利用单层垂直上下文的局部模糊检测优化算法.首先提取图像块重尾分布、峰度、功率谱、线性滤波等模糊特征,然后使用贝叶斯法学习模型,计算后验概率作为初步估计模糊响应,最后本文提出将邻近像素点的模糊响应信息作为上下文更新像素点自身响应信息,增加上下文支撑域的尺寸以更充分的考虑周围信息,使用一个相互垂直的一维上下文以减小计算量,从而构造新的能量函数进行全局优化,通过最小化能量函数得到最终的模糊响应.实验表明,本文算法能有效检测图像的局部模糊,并提高检测的时间效率.

Local Blur Detection Optimization Algorithm

The existing blur detection algorithm based on features has to calculate local features many times in order to optimize the blur response. Complicated adjacency relationship leads to large computing tasks and low efficiency. To solve the time problem, this paper proposes a local blur detection optimization algorithm with the perpendicular context in single scale. Firstly, we calculate features such as local heavy-tailedness feature, kurtosis feature, local power spectrum feature and local filters, Then naive Bayesian classifier is used to combine these features. Finally, this paper takes the blur response information of adjacent pixels as context to update the pixel itself, increases support region of context to take the information around the pixel into account more adequately, uses one-dimensional perpendicular context to reduce the amount of calculation, forms new energy function, and obtains the final blur response by minimizing the energy function. Experimental results show that the modified algorithm can detect local blur effectively and improve time efficiency.