Image copy detection HDU - 3523 最小费用流

本文介绍了一种用于检测图像副本的系统,该系统通过将图像划分为等大小块并计算其平均值来创建一个等级矩阵,进而实现对数字媒体复制的高效追踪及版权保护。
The success of the Internet and cost-effective digital storage device has made it possible to replicate, transmit, and distribute digital content in an effortless way. Thus, the protection of intellectual Property right (IPR) has become a crucial legal issue. Detecting copies of digital media(images, audio and video) is a basic requirement for IPR protection (or copyright protection).The applications of copydetection include usage tracking and copyright violation enforcement.
For those above purposes, the image copy detection system came out.It aggregated all the images which were viewed as a copy of original image.In this system,an image is partitioned into m×n equalsized blocks,which makes the system independent of input image sizes,and an m×n sub-image is calculated by taking the average value of each block(see Figure.1(b)), This array is converted to a rank matrix as shown in Fig.1(c).Suppose that the intensity values in Fig.1(b) are changed in the copied image so that its sub-image has values:{{30, 60, 40}, {70, 90, 110}, {50, 100, 80}}. Nevertheless,its rank matrix is identical to that shown in Fig. 1(c) and thus perfect matching with original image can be achieved.



Fig.1 (a)An image is divided into m×n blocks (3×3 in this example),(b)average values of blocks, and(c)rank matrix of (b). Let T and Q represent test image and original image,N represent the matrix size;there exists N tuples (t1,q1),…,( tn,qn),…,( tN,qN)(the order of the rank matrix:from left to right and from top to bottom).Now we define D(T,Q) = measures the distance between the two images(it’s obvious that if D(T,Q) gets smaller while the probability of the test image is considered as a copy of original image by the system becomes larger). Since there are M original images in the image copy detection system(Q1…QM). And the distance between T and some original images is given by D(T,Q1…QM)= To make it simple,we want to find an image which owns least D(T,Q1…QM).

Input
The first line of input should give the number of cases, T (at most 100). T test cases follow. The first line of each test case contains two integer n (1 ≤ n ≤ 100) and m(1 ≤ m ≤ 100) indicating the size of rank matrix and number of original images. The following m lines each contains a sequence of n different integers denotes the rank matrix.
Output
For each test case, output one line containing "Case #x: y", where x is the case number (starting from 1) and y is the least distance D(T,Q1…QM).
Sample Input
2
3 2
1 2 3
1 3 2
9 3
1 4 2 5 7 9 3 8 6
2 1 4 5 9 3 7 6 8
9 8 7 6 5 4 3 2 1
Sample Output
Case #1: 2
Case #2: 58




#include<cstdio>
#include<cstring>
#include<algorithm>
#include<iostream>
#include<cmath>
#include<vector>
#include<queue>
using namespace std;
typedef long long ll;
const int inf = 0x3f3f3f3f;
const int M = 1e3 + 10;
typedef pair<int,int > P;//first保存最短距离,second保存顶点编号
//表示边的结构体(终点、容量、费用、反向边)
struct edge
{
    int to,cap,cost,rev;
    //edge();
   // edge(int a,int b,int c,int d){to=a;cap=b;cost=c;rec=d;};
};
int t;
int V;//顶点数
int n,m;
vector<edge>G[M];
int h[M],dis[M];//图的势(指的是每个顶点赋予的一个标号h(v)) 最短距离
int prevv[M],preve[M]; //图的最短路中的前驱节点和对应的边
//增加一条从a到b容量为c费用为d的一条边
void add_edge(int a,int b,int c,int d)
{
    G[a].push_back(edge{b,c,d,G[b].size()});
    G[b].push_back(edge{a,0,-d,G[a].size()-1});
}
//求解从s到t流量为F的最小费用流
// 如果没有流量为F 的流,输出为-1
int min_cost_flow(int st,int en,int f){
    int ans=0;
    memset(h,0,sizeof(h));
    while(f>0){
        priority_queue<P,vector<P>,greater<P> >que;
        for(int i=0;i<M;i++) dis[i]=inf;
        dis[st]=0;que.push(P(0,st));
        while(!que.empty()){
            P p=que.top();que.pop();
            int v=p.second;
            if(dis[v]<p.first) continue;
            for(unsigned int i=0;i<G[v].size();i++){
                edge &e=G[v][i];
                if(e.cap>0&&dis[e.to]>dis[v]+e.cost+h[v]-h[e.to]){
                    dis[e.to]=dis[v]+e.cost+h[v]-h[e.to];
                    prevv[e.to]=v;
                    preve[e.to]=i;
                    que.push(P(dis[e.to],e.to));
                }
            }
        }
        if(dis[en]==inf) return -1;
        for(int i=0;i<M;i++) h[i]+=dis[i];
        int d=f;
        for(int i=en;i!=st;i=prevv[i]){
            d=min(d,G[prevv[i]][preve[i]].cap);
        }
        f-=d;
        ans+=d*h[en];
        for(int i=en;i!=st;i=prevv[i]){
            edge &e=G[prevv[i]][preve[i]];
            e.cap-=d;
            G[i][e.rev].cap+=d;
        }
    }
    return ans;
}
int num[110][110],tmp[110][110];
int main()
{
    scanf("%d",&t);
    int cas=1;
    while(t--)
    {
        scanf("%d%d",&n,&m);
        for(int i=0;i<M;i++)
            G[i].clear();
        for(int i=1;i<=m;i++)
            for(int j=1;j<=n;j++)
            scanf("%d",&num[i][j]);
        for(int i=1;i<=n;i++)
            add_edge(0,i,1,0);
        for(int i=1;i<=n;i++){
            for(int j=1;j<=n;j++){
                int sum=0;
                for(int k=1;k<=m;k++) sum+=abs(num[k][j]-i);
                tmp[j][i]=sum;
            }
        }
        for(int i=1;i<=n;i++){
            for(int j=1;j<=n;j++){
                add_edge(i,n+j,1,tmp[j][i]);
            }
        }
        for(int i=1;i<=n;i++) add_edge(n+i,2*n+1,1,0);
        int ans=min_cost_flow(0,2*n+1,n);
        printf("Case #%d: %d\n",cas++,ans);
    }
    return 0;
}


「LLM那些事」系列第 4 篇《上下文窗口的边界》,文章连接:https://blog.csdn.net/houwenjin/article/details/163999753。 演示什么:在「预测」Sheet 的黄色格子里输入一句话(默认「来泡一杯」),四个「模型」——分别只统计最后 1 / 2 / 3 / 4 个字的 n-gram 查表——同时预测下一个字。同一个输入,看的上下文越长,候选越少、预测越确定: ┌────────────────┬──────────┬───────────────┬──────┐ │ 只看最后几个字 │ 用的前缀 │ 候选下一字数 │ 预测 │ ├────────────────┼──────────┼───────────────┼──────┤ │ 1 个 │ 杯 │ 3(茶/子/水) │ 模糊 │ ├────────────────┼──────────┼───────────────┼──────┤ │ 2 个 │ 一杯 │ 2(茶/水) │ 收窄 │ ├────────────────┼──────────┼───────────────┼──────┤ │ 3 个 │ 泡一杯 │ 1(茶) │ 确定 │ ├────────────────┼──────────┼───────────────┼──────┤ │ 4 个 │ 来泡一杯 │ 1(茶) │ 确定 │ └────────────────┴──────────┴───────────────┴──────┘
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