基于特征显著值归一化与位置加权的FT算法

Frequency-Tuned Salient Region Detection Algorithm Based on Feature Saliency Normalization and Position Weighting

频率调谐(FT)显著区域检测算法在背景复杂和图像显著区域比较大时检测效果不理想。针对上述问题,对FT算法进行了改进,提出了一种基于特征显著值归一化与位置加权的频率调谐显著区域检测算法(FTFP)。该算法主要在FT算法的基础上进行了图像分块、Lab颜色特征显著值的分别归一化和位置加权处理。实验结果表明,FTFP算法在显著性检测视觉效果、准确率与查全率、对噪声图像的检测上都优于FT算法,综合性能突出。

Frequency-tuned（ FT） detection algorithms are not ideal in complex backgrounds and various large salient regions of an image. In view of the above problems,the FT algorithm was improved,and a new frequency-tuned salient region detection algorithm based feature saliency normalization and position weighting was proposed（ FTFP）. The algorithm divided the image into blocks,and normalized the Lab color space characteristic saliency values,and was weighted by position based on the FT algorithm. The experimental results show that the visual effect,accuracy rate and recall,the detection of image noise of algorithm FTFP in saliency detection are better than that of the original FT algorithm,and have outstanding performance.