For the image saliency detection in mass visual data processing,we propose a novel multiple-feature information-based method. First,on the basis of CIE Lab colour space and spatial position information of pixels,the method selects k-means algorithm to cluster image pixels.In initialising the centre it employs the regular hexagon for seed selection according to honeycomb conjecture.Then it analyses the selected multiple image features with global contrast and local contrast methods,and derives eight feature maps through computation.At last, it integrates these eight feature maps to obtain the initial saliency map,and then gets the final saliency map using threshold method.By improving k-means algorithm the proposed method realises excellent image clustering results to further analyse and process multiple image features.Taking into account the global and local contrasts,the proposed method reasonably integrates the low-level and middle-level features of image according to some vital principles,e.g.,the contrast and the key regional focus,etc.,and deals with the problems comprehensively and efficiently.Experimental results show that the proposed method achieves superior performances according to the overall assessments from both the subjective and objective aspects.%针对海量视觉数据处理中的图像显著性检测问题，提出一种基于多重特征信息的新型方法。该方法首先根据像素的CIE Lab 颜色空间和空间位置信息选用 k-means 算法对图像像素聚类，在初始化中心时根据蜂窝原理使用正六边形进行选种。然后用全局对比和局部对比方法分析选取的多重图像特征，并计算得到八种特征图。最后对八种特征图融合得到初始显著性图，再用阈值法得到最终的显著性图。该方法通过改进 k-means 算法实现良好的图像聚类以进一步分析、处理图像特征，并依据对比度、关键区域聚焦等重要原理将图像底层特征和中层特征合理融合，兼顾全局对比和局部对比，处理问题全面而高效。实验结果表明，从主观和客观两方面进行整体评估，该方法都达到了优越的性能。
展开▼
