颜色空间转换耦合特征图的显著性检测算法

为解决当前视觉显著性检测技术忽略图像的全局与局部特征的联系,使其对复杂图像的检测准确度不佳的问题,设计颜色空间转换图耦合Ripplet变换的视觉显著性检测算法。将图像转换到RGBYI空间,并计算R与G、B与Y分量的颜色差异;引入Ripplet变换,对图像进行分解,获取相应的变换系数;借助逆Ripplet变换,形成特征图;基于概率密度函数,联合特征图,计算全局显著图;利用逐像素相似度估算像素的信息熵,获取图像的局部显著图;通过组合局部与全局显著图,形成最终的显著图,完成目标检测。实验结果表明,与当前显著性检测技术相比,所提技术具有更好的检测效果,呈现出更为理想的接收机工作特性曲线。

Saliency detection algorithm based on color space conversion and feature map

To solve the problem of poor detection effects of complex images induced by ignoring the relation between the global and local features in current visual saliency detection technique, the image visual saliency detection algorithm based on multi-feature map and Ripplet transform was proposed. The image was transformed into RGBYI space, and color difference between the R and G components as well as that between B and Y components were calculated. The image was decomposed by introducing the Ripplet transform to obtain the corresponding transform coefficients. The feature map was formed by inverse Ripplet transform. The global saliency was calculated based on the probability density function and feature map. The local saliency map of the image was obtained by using the pixelwise similarity to compute the information entropy of pixels. The final saliency region was formed by combining local and global saliency map to achieve target detection. Experimental results show that this technique has better detection results and better receiver operating characteristic curve characteristics compared with the current saliency detection technology.