纹理粗糙度在红外图像显著性检测中的应用

提出了基于纹理粗糙度的红外图像显著性检测算法,以解决红外图像对比度低,目标显著性检测难的问题。首先,研究了Tamura的粗糙度原理,对粗糙度进行分析和评价,提出了新的粗糙度计算方法。然后,将图像分解为超级像素集合,并计算超级像素的最大平均强度差;利用最大平均强度差定义超级像素的最佳尺度,作为纹理粗糙度的度量。最后,将超级像素区域均匀外延,利用粗糙度的局部对比度和灰度信息度量红外图像的显著性。通过实验验证了本文算法的有效性,结果表明:在10%的噪声水平下,本文粗糙度保持不变,粗糙度特征图一致性较好,而Tamura的粗糙度特征图中杂点明显增多。与其它显著性检测算法对比,本文算法击中率最高,为0.752。该算法挖掘了红外图像的纹理粗糙度特征,为红外图像显著性检测提供了新的特征选择。

Application of texture coarseness in saliency detection of infrared image

A saliency detection algorithm for infrared images based on texture coarseness was proposed to detect the saliency of targets owing to a low image contrast. Firstly, Tamura's principle of calculating coarseness was researched, and a new method to calculate the coarseness was presented by analysis and evaluation of the coarseness. Then the image was decomposed into a set of super pixels and the maximum mean intensity difference of the super pixels was calculated. Furthermore, the best scale of super pixels was defined by using maximum mean intensity difference to be a measure of the texture coarseness. Finally, the region of every super pixel was expanded isotropically and the saliency of infrared image was measured based on the local contrast and grey information of the texture coarseness with the weight of intensity. The effectiveness of algorithm was verified. Results show that coarseness based on the proposed method remains unchanged under a noise level of 10% and the homogeneity is better in the feature map of coarseness. Meanwhile, there are many miscellaneous points in Tamura's feature map of coarseness. Compared with other methods of saliency detection for infrared images, the proposed algorithm has the highest hit rate, up to 0.752. The algorithm exploits the feature of texture coarseness, and provides a new selection method for the saliency detection of infrared images.