diff --git a/modules/cudaarithm/src/cuda/absdiff_mat.cu b/modules/cudaarithm/src/cuda/absdiff_mat.cu
index da475563e25..1df17751daf 100644
--- a/modules/cudaarithm/src/cuda/absdiff_mat.cu
+++ b/modules/cudaarithm/src/cuda/absdiff_mat.cu
@@ -49,6 +49,7 @@
 #else
 
 #include "opencv2/cudev.hpp"
+#include <climits>
 
 using namespace cv::cudev;
 
@@ -77,6 +78,16 @@ namespace
         }
     };
 
+    template <> struct AbsDiffOp1<int> : binary_function<int, int, int>
+    {
+        __device__ __forceinline__ int operator ()(int a, int b) const
+        {
+            const long long diff = static_cast<long long>(a) - static_cast<long long>(b);
+            const long long absDiff = diff < 0 ? -diff : diff;
+            return static_cast<int>(absDiff > INT_MAX ? INT_MAX : absDiff);
+        }
+    };
+
     template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
     };
diff --git a/modules/cudaarithm/test/test_element_operations.cpp b/modules/cudaarithm/test/test_element_operations.cpp
index 46e156d993e..537eb9dde2d 100644
--- a/modules/cudaarithm/test/test_element_operations.cpp
+++ b/modules/cudaarithm/test/test_element_operations.cpp
@@ -1404,6 +1404,23 @@ CUDA_TEST_P(AbsDiff, Array)
     }
 }
 
+CUDA_TEST_P(AbsDiff, IntMin)
+{
+    if (depth != CV_32S)
+        return;
+
+    cv::Mat src1 = (cv::Mat_<int>(1, 2) << std::numeric_limits<int>::min(), 0);
+    cv::Mat src2 = (cv::Mat_<int>(1, 2) << 0, std::numeric_limits<int>::min());
+
+    cv::Mat expected = (cv::Mat_<int>(1, 2) << std::numeric_limits<int>::max(),
+                                                   std::numeric_limits<int>::max());
+
+    cv::cuda::GpuMat dst;
+    cv::cuda::absdiff(loadMat(src1), loadMat(src2), dst);
+
+    EXPECT_MAT_NEAR(expected, dst, 0.0);
+}
+
 CUDA_TEST_P(AbsDiff, Scalar)
 {
     cv::Mat src = randomMat(size, depth);