针对传统非DCT 域图像显著区域检测方法无法直接应用于压缩图像的问题, 提出一种基于DCT 系数的图像显著区域检测方法. 首先利用DCT 所得系数提取图像块的亮度、颜色和纹理特征; 然后通过分析特征向量分布规律对频率最低的部分特征向量进行替换, 并计算特征向量全局对比度; 最后将构建的高斯系数矩阵与所得对比度结果融合, 从而完成图像显著区域检测. 仿真实验结果表明, 与现有的基于压缩域的显著检测方法相比, 该方法可获得更好的显著区域检测结果, 并且有着较高的检测效率.
At present, network transmission is an important component of image applications, such as online im-age retrieval, etc. However, the conventional image salient region detection methods based on uncompressed do-main can not be directly applied to the compressed image. To solve the above problem, a new image saliency de-tection method based on DCT coefficients is presented. At first, luminance, color and texture features of the im-age blocks were extracted based on the coefficients that obtained from DCT transform. Then distribution of fea-ture vectors was analyzed statistically, and partial feature vectors with lower frequency were replaced, and the global contrast of feature vectors was calculated based on the replacement results. Finally, the Gaussian coeffi-cient matrices and the previous results were fused, thus the salient region detection task was finished. The simula-tion results demonstrate that the proposed method can obtain better saliency detection results and higher detection efficiency compared to existing method based on compressed domain.