多特征融合的低景深图像前景提取算法

Foreground Extraction from Low Depth-of-field Images Based on Colour-texture and HOS Features

针对低景深(Low depth-of-field,DOF)图像,提出了一种融合纹理、颜色和高阶统计量(Higher-order statistics,HOS)特征的聚焦前景提取方法.首先,根据相似性最大化原则,通过迭代获得纹理和颜色特征的优化权重,实现低景深图像的区域分割.然后,根据优化权重值计算颜色空间上的加权HOS值,并结合区域归属前景的划分策略,实现低景深图像的前景提取.实验结果表明,该算法可以同时取得较高的主观和客观评价效果.

This paper presents a new algorithm for extracting foreground objects from low depth-of-field （DOF） images using texture, color and high-order statistics （HOS） features. Firstly, an algorithm with automatic weight optimization is designed to segment DOF images according to the principle of maximum similarity. The foreground of DOF images is then extracted based on the weighted HOS and a strategy for foreground region classification. Simulation results demonstrate that the proposed algorithm achieves satisfactory result both subjectively and objectively.