#include "bits/stdc++.h"

#include <stdio.h>
#include <math.h>

#include "atcoder/all"
#include <float.h>

#include <boost/multiprecision/cpp_int.hpp>
using namespace boost::multiprecision;

using namespace std;
using namespace atcoder;

using ll = long long;
using ull = unsigned long long;
using mint = modint998244353;
using mint2 = modint1000000007;
const double eps = 1e-9;

#define REP(i, n) for (ll i = 0; i < ll(n); ++i)
#define REPI(i, init, n) for (ll i = init; i < ll(n); ++i)
#define REPD(i, init) for (ll i = init; i >=0; --i)
#define REPDI(i, init, n) for (ll i = init; i >=n; --i)
#define REPE(i, c) for (const auto& i : c)
#define TCASE() ll _ttt; cin >> _ttt; while(_ttt--)

using vl = vector<ll>; using vvl = vector<vl>; using vvvl = vector<vvl>; using dl = deque<ll>;
#define VLI(a, n) vector<ll> a; a.assign(n, 0); for(auto& x : a) cin >> x;
#define DLI(a, n) deque<ll> a; a.assign(n, 0); for(auto& x : a) cin >> x;
#define VSI(a, n) vector<string> a; a.assign(n, ""); for(auto& x : a) cin >> x;
#define VVLI(a, n, m, init) vector<vector<ll>> a(n); for(auto& x : a) x.assign(m, init);
using vd = vector<double>; using pl = pair<ll, ll>; using vpl = vector<pl>;

struct uv { ll u; ll v; ll c; };
#define VUVI(a, n) vector<uv> a; a.assign(n, {0, 0, 1}); for(auto& x : a) {cin >> x.u >> x.v; x.u--; x.v--;}
#define VUVCI(a, n) vector<uv> a; a.assign(n, {0, 0, 0}); for(auto& x : a) {cin >> x.u >> x.v >> x.c; x.u--; x.v--;}
vvl to_edge(const ll n, const vector<uv>& v) { vvl ret(n); for (auto& x : v) ret[x.u].push_back(x.v); return ret; }
vvl to_edge_d(const ll n, const vector<uv>& v) { vvl ret(n); for (auto& x : v) { ret[x.u].push_back(x.v); ret[x.v].push_back(x.u); } return ret; }

template <class T = long long> using pql = priority_queue<T>;
template <class T = long long> using pqg = priority_queue<T, vector<T>, greater<T>>;
using vm = vector<mint>; using vvm = vector<vm>;

ostream& operator<<(ostream& os, const mint& v) { os << v.val(); return os; }
void yn(bool f) { std::cout << (f ? "Yes" : "No") << endl; };
template<class T> void ov(const T& v) { for (auto it = v.begin(); it != v.end(); it++) { if (it != v.begin()) cout << " "; cout << *it; } };
template<> void ov(const vm& v) { for (auto it = v.begin(); it != v.end(); it++) { if (it != v.begin()) cout << " "; cout << it->val(); } };

const ll llhuge = 1LL << 62; //十分でかいが多少足しても溢れない数値
auto gmax(const auto& a, const auto& b) { return a > b ? a : b; };
auto gmin(const auto& a, const auto& b) { return a < b ? a : b; };
bool chmin(auto& a, const auto& b) { const auto olda = a; a = gmin(a, b); return olda != a; }
bool chmax(auto& a, const auto& b) { const auto olda = a; a = gmax(a, b); return olda != a; }
void sort(auto& v) { std::ranges::sort(v); };
void rev(auto& v) { std::ranges::reverse(v); };
template<class T> auto p_sum(const T& v) { T ret(v.size()); std::partial_sum(begin(v), end(v), ret.begin()); return ret; };
template<class T, class K = T::key_type, class V = T::mapped_type> [[nodiscard]] V myat(const T& mp, const K& key) { return mp.find(key) == mp.end() ? V() : mp.at(key); }
long long ceil(const long long a, const long long b) { return (a + b - 1) / b; }

//二分探索でf(t)を満たす整数tを返す
//ng < ok ならtは最小値 ng > ok ならtは最大値
template <class F>
ll bs(ll ng, ll ok, F f) {
	while (abs(ok - ng) > 1) {
		ll mid = (ok + ng) / 2;

		if (f(mid)) ok = mid;
		else ng = mid;
	}
	return ok;
}

int main() {
	cin.tie(nullptr);
	ios::sync_with_stdio(false);

	ll n, m, k;
	cin >> n >> m >> k;

	vvl a(n, vl(m));
	REP(i, n) REP(j, m) cin >> a[i][j];

	ll ans = bs(1000000001, 0, [&](ll x) {
		if (x == 1000000001) return false;

		vvl len(n, vl(m, llhuge));

		deque<pair<pl, ll>> q;
		q.push_back({ { 0, 0 }, 0 });
		//01BFS
		while (!q.empty()) {
			auto [p, ln] = q.front();
			auto [i, j] = p;
			q.pop_front();

			if (len[i][j] <= ln) continue;
			len[i][j] = ln;

			REP(k, 4) {
				ll dx = (k == 0) ? 1 : (k == 1) ? -1 : 0;
				ll dy = (k == 2) ? 1 : (k == 3) ? -1 : 0;
				ll ni = i + dx;
				ll nj = j + dy;
				if (ni < 0 || ni >= n || nj < 0 || nj >= m) continue;

				if (a[i][j] >= x) {
					q.push_front({ { ni, nj }, ln });
				}
				else {
					q.push_back({ { ni, nj }, ln + 1 });
				}
			}
		}

		if (a[n-1][m-1] < x) len[n - 1][m - 1]++;
		return len[n-1][m-1] <= k;
	});

	cout << ans << endl;

}