基于余弦相似度的Graph Cuts序列图像分割算法

Graph Cuts Sequence Image Segmentation Algorithm Based on Cosine Similarity

传统Graph Cuts算法容易出现漏分割、误分割现象,分割效果有待提高,且需要人工交互分割效率不高。针对传统算法的不足,提出基于余弦相似度的Graph Cuts序列分割图像算法。使用最大余弦相似度构造能量函数的区域项,计算超像素与种子聚类区域的最大余弦相似度。使用余弦相似度和颜色相似度来构造边界项,计算邻域超像素的颜色和相对距离特征相似度。将该算法与传统Graph Cuts算法进行分割结果比较精确率和召回率等均有提高,漏报率降和虚报率明显降低。算法应用于序列图像分割,减少人工交互,提高分割效率。

Traditional Graph Cuts segmentation algorithm may bring leakage segmentation and error segmentation. The segmentation effect of the algo- rithm needs to be improved, and the algorithm needs human interaction with low segmentation efficiency. In order to improve the shortcom- ings of traditional algorithms, proposes the Graph Cuts algorithm of sequence image segmentation based on cosine similarity. Constructs the region term of the energy function by using the maximum cosine similarity, and calculates the maximum cosine similarity between the su- per-pixel and the seed clustering region. Constructs the boundary term of the energy function by using the cosine similarity and color simi- larity, and calculates the color and the relative distance feature similarity of neighborhood super-pixel. This algorithm is compared with the traditional Graph Cuts algorithm, the accuracy and recall rate are improved, the false positive rate and the false negative rate is reduced ob- viously. Applies this algorithm to sequence image segmentation, which reduces the manual interaction and improves the segmentation effi- ciency.