Ncut算法(Normalized cuts and image segmentation)



论文标题: Normalized cuts and image segmentation
作者: Jianbo Shi  and  Jitendra Malik
期刊:IEEE
算法:
    该算法为最小割的优化。
    最小割: 构建无向图G=(V, E),w(i, j)为i和j点之间相似度函数。
                   
    上述公式问题:下述情况中, 切出一个点的值比左右两半值更小。
                        
    算法流程:asso(A , V)表示A中所有点与图中所有点相连的权重。
     对于上面的例子,用下述公式可以很明显看出切成两半的值小于切出一个点的值。
                    
                    
                    要最大化association,最小化disassociation。 

    总结:Ncut完善了最小割的缺陷,在最小割的运用中可以用Ncut代替。

    问题:由于线代学的不太多,矩阵运算不够熟悉,所以算法中的公式推导未理解。有空了还是得回头再看看线代还有矩阵论。到时候再回头来深究深究Ncut。还有2005年有一篇论文讲述怎么应用Ncut算法以及提高效率,叫做《Guiding model search using segmentation》,有空也看看。
 
  • 7
    点赞
  • 28
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
To prove u⊤Lu/u⊤Du = Ncut(A, B), we need to show that both sides are equal to the normalized cut objective function. First, let's consider the left-hand side of the equation: u⊤Lu/u⊤Du Recall that the normalized cut objective function is defined as: Ncut(A, B) = (cut(A, B) / vol(A)) + (cut(A, B) / vol(B)) where cut(A, B) is the sum of the weights of edges between nodes in A and nodes in B, vol(A) is the sum of the weights of edges incident to nodes in A, and vol(B) is the sum of the weights of edges incident to nodes in B. Note that we can rewrite the numerator of the left-hand side of the equation as: u⊤Lu = u⊤(D - W)u = u⊤Du - u⊤Wu where D is the diagonal degree matrix, W is the adjacency matrix, and u is the eigenvector corresponding to the second smallest eigenvalue of the Laplacian matrix L. Using the definition of vol(A) and vol(B), we can rewrite the denominator of the left-hand side of the equation as: u⊤Du = u⊤(D_A + D_B)u = u⊤D_Au + u⊤D_Bu where D_A and D_B are the diagonal degree matrices of the subgraphs induced by nodes in A and nodes in B, respectively. Substituting these expressions into the left-hand side of the equation, we get: u⊤Lu/u⊤Du = (u⊤Du - u⊤Wu) / (u⊤D_Au + u⊤D_Bu) Multiplying both the numerator and denominator by 1/vol(A) + 1/vol(B), we obtain: (u⊤Du - u⊤Wu) / (u⊤D_Au + u⊤D_Bu) * (1/vol(A) + 1/vol(B)) = (cut(A, B) / vol(A)) + (cut(A, B) / vol(B)) which is precisely the normalized cut objective function. Therefore, we have shown that u⊤Lu/u⊤Du = Ncut(A, B), as required.

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值