【摘要】 为提高图像显著性检测的准确性与有效性,在CIE Lab颜色空间内,通过模拟生物视觉神经元的中央-周围运算,提出一种基于稀疏表示与自信息的快速显著性检测方法。对原始输入图像的特征图像进行稀疏量化,计算该稀疏量化图像各像素点的自信息值,根据各像素点的自信息值进行显著性检测。实验结果表明,与GBVS、AIM和ITTI模型相比,该方法 AUC值分别提高了15%、17%和20%,平均耗时则分别降低了93%、95%和92%,验证了该方法能够准确快速地检测图像显著性。更多还原

【Abstract】 To increase the accuracy and efficiency of image saliency detection,through mimicking the center-surround(C-S)operations of biological vision neuron,a fast saliency detection method based on sparsity and self-information was proposed in the CIE Lab color space.The image features of original input image were sparsely quantified,and the self-information of each pixel in the sparse matrix was calculated.Subsequently,saliency detection was implemented according to the self-information of each pixel.Experimental results show that,comparing with the GBVS,AIM and ITTI models,the areas under the receiver operating characteristic curve(AUC)of the proposed method increase 15%,17% and 20%,respectively.Correspondingly,the average time-consuming decreases 93%,95% and 92%,respectively.The proposed method can detect the saliency more accurately and efficiently.更多还原