基于图论的视觉注意模型

为了模拟人眼的视觉搜索结果,提出了一种基于图论的自底向上的视觉注意计算模型.在计算显著图的过程中,构造了马尔可夫矩阵并计算图中每个顶点的度.由马尔可夫矩阵和度的计算提取出图像在颜色、亮度和方向3个特征方面显著的区域,最后生成显著图.实验仿真结果表明:与部分的视觉注意模型相比,本文模型具有目标检测能力强和计算量小的优点

Research of visual attention model based on graph

To simulate the visual search result of human, a new bottom-up visual attention model was proposed based on the graph theory. In textual model, the procession of getting salience map consisted of two steps. The first step constracted the markovian marix.The second one accounted the degree of every vertex in graph, which highlighted the salient area of maps on certain feature channel including colors, intensity and orientations. Experimental results showed that this model is more effective for target detection and has less calculating works than some others visual attention models.