光学 精密工程, 2014, 22 (4): 1012, 网络出版: 2014-05-06
融合对比度和分布性的图像显著性区域检测
Detection of salient maps by fusion of contrast and distribution
视觉显著性 RGB空间 LAB空间 对比度 分布性 主成分分析 visual saliency RGB space LAB space contrast distribution Principal Component Analysis(PCA)
摘要
单独基于对比度的显著性检测方法由于忽略了特征的空间分布, 且只在RGB空间或LAB空间下单独进行计算, 故实验结果不理想。本文提出了结合RGB和LAB两种特征空间并融合了对比度和分布性的图像显著性区域检测算法。该算法首先提取图像分块在RGB空间和LAB空间下的原始特征并进行组合, 在主成分分析(PCA)降维的基础上自动选择有效特征; 然后计算图像分块的对比度和分布性, 融合对比度特征和分布性特征实现对原始图像的显著性区域提取。实验结果显示, 本文算法的平均准确率为0.821 7, 平均召回率为0.692 5, 综合指标F值达0.787 8。计算的显著性区域的效果比在RGB空间或LAB空间下单独基于对比度的计算方法有明显改善, 相比其他检测方法更加准确, 符合人眼的观测结果, 均匀突出了显著性区域。
Abstract
The existing saliency detection algorithm can not obtain an ideal result because the contrast based method ignores the specific spatial distribution and calculates only in a RGB space or a LAB space. An algorithm of salient region detection based on the fusion of contrast and distribution under the combination of RGB space and LAB space was proposed. By this method, original image patches in the RGB space and the LAB space were extracted and combined, and the effective features were automatically selected based on Principle Component Analysis(PCA) dimensionality reduction. The contrast and distribution of image patches were calculated in the reduced dimensional space and finally were fused to extract the saliency region. Experimental results show that the precision ratio, the recall ratio and overall F-measure of the proposed detection are 0.821 7, 0.692 5 and 0.787 8, respectively. The effect of the proposed algorithm is more improved than the two algorithms based on the contrast in the RGB space or the LAB space alone. This method is more accurate and is more in line with the human eye observation results, uniformly highlighting the whole salient areas.
张颖颖, 张帅, 张萍, 卢成. 融合对比度和分布性的图像显著性区域检测[J]. 光学 精密工程, 2014, 22(4): 1012. ZHANG Ying-ying, ZHANG Shuai, ZHANG Ping, LU Cheng. Detection of salient maps by fusion of contrast and distribution[J]. Optics and Precision Engineering, 2014, 22(4): 1012.