diff --git a/src/nvimage/Filter.cpp b/src/nvimage/Filter.cpp
index e650a21..db06d80 100644
--- a/src/nvimage/Filter.cpp
+++ b/src/nvimage/Filter.cpp
@@ -283,7 +283,6 @@ void GaussianFilter::setParameters(float variance)
 
 
 
-/// Ctor.
 Kernel1::Kernel1(const Filter & f, int iscale, int samples/*= 32*/)
 {
     nvDebugCheck(iscale > 1);
@@ -312,13 +311,12 @@ Kernel1::Kernel1(const Filter & f, int iscale, int samples/*= 32*/)
     }
 }
 
-/// Dtor.
 Kernel1::~Kernel1()
 {
     delete m_data;
 }
 
-/// Print the kernel for debugging purposes.
+// Print the kernel for debugging purposes.
 void Kernel1::debugPrint()
 {
     for (int i = 0; i < m_windowSize; i++) {
@@ -328,13 +326,18 @@ void Kernel1::debugPrint()
 
 
 
-/// Ctor.
 Kernel2::Kernel2(uint ws) : m_windowSize(ws)
 {
     m_data = new float[m_windowSize * m_windowSize];
 }
 
-/// Copy ctor.
+Kernel2::Kernel2(uint ws, const float * data) : m_windowSize(ws)
+{
+    m_data = new float[m_windowSize * m_windowSize];
+
+    memcpy(m_data, data, sizeof(float) * m_windowSize * m_windowSize);
+}
+
 Kernel2::Kernel2(const Kernel2 & k) : m_windowSize(k.m_windowSize)
 {
     m_data = new float[m_windowSize * m_windowSize];
@@ -344,13 +347,12 @@ Kernel2::Kernel2(const Kernel2 & k) : m_windowSize(k.m_windowSize)
 }
 
 
-/// Dtor.
 Kernel2::~Kernel2()
 {
     delete m_data;
 }
 
-/// Normalize the filter.
+// Normalize the filter.
 void Kernel2::normalize()
 {
     float total = 0.0f;
@@ -364,7 +366,7 @@ void Kernel2::normalize()
     }
 }
 
-/// Transpose the kernel.
+// Transpose the kernel.
 void Kernel2::transpose()
 {
     for(uint i = 0; i < m_windowSize; i++) {
@@ -374,7 +376,7 @@ void Kernel2::transpose()
     }
 }
 
-/// Init laplacian filter, usually used for sharpening.
+// Init laplacian filter, usually used for sharpening.
 void Kernel2::initLaplacian()
 {
     nvDebugCheck(m_windowSize == 3);
@@ -392,7 +394,7 @@ void Kernel2::initLaplacian()
 }
 
 
-/// Init simple edge detection filter.
+// Init simple edge detection filter.
 void Kernel2::initEdgeDetection()
 {
     nvCheck(m_windowSize == 3);
@@ -401,7 +403,7 @@ void Kernel2::initEdgeDetection()
     m_data[6] = 0; m_data[7] = 0; m_data[8] = 0;
 }
 
-/// Init sobel filter.
+// Init sobel filter.
 void Kernel2::initSobel()
 {
     if (m_windowSize == 3)
@@ -460,7 +462,7 @@ void Kernel2::initSobel()
     }
 }
 
-/// Init prewitt filter.
+// Init prewitt filter.
 void Kernel2::initPrewitt()
 {
     if (m_windowSize == 3)
@@ -486,7 +488,7 @@ void Kernel2::initPrewitt()
     }
 }
 
-/// Init blended sobel filter.
+// Init blended sobel filter.
 void Kernel2::initBlendedSobel(const Vector4 & scale)
 {
     nvCheck(m_windowSize == 9);
@@ -609,7 +611,7 @@ PolyphaseKernel::~PolyphaseKernel()
 }
 
 
-/// Print the kernel for debugging purposes.
+// Print the kernel for debugging purposes.
 void PolyphaseKernel::debugPrint() const
 {
     for (uint i = 0; i < m_length; i++)
diff --git a/src/nvimage/Filter.h b/src/nvimage/Filter.h
index bb9ca7d..6ced86b 100644
--- a/src/nvimage/Filter.h
+++ b/src/nvimage/Filter.h
@@ -166,6 +166,7 @@ namespace nv
     {
     public:
         Kernel2(uint width);
+        Kernel2(uint width, const float * data);
         Kernel2(const Kernel2 & k);
         ~Kernel2();
 
diff --git a/src/nvimage/FloatImage.cpp b/src/nvimage/FloatImage.cpp
index b25c7d6..f61c9e7 100644
--- a/src/nvimage/FloatImage.cpp
+++ b/src/nvimage/FloatImage.cpp
@@ -943,6 +943,26 @@ FloatImage * FloatImage::resize(const Filter & filter, uint w, uint h, uint d, W
 }
 
 
+void FloatImage::convolve(const Kernel2 & k, uint c, WrapMode wm)
+{
+    AutoPtr<FloatImage> tmpImage = clone();
+
+    uint w = m_width;
+    uint h = m_height;
+    uint d = m_depth;
+
+    for (uint z = 0; z < d; z++)
+    {
+        for (uint y = 0; y < h; y++)
+        {
+            for (uint x = 0; x < w; x++)
+            {
+                pixel(c, x, y, 0) = tmpImage->applyKernelXY(&k, x, y, z, c, wm);
+            }
+        }
+    }
+}
+
 
 /// Apply 2D kernel at the given coordinates and return result.
 float FloatImage::applyKernelXY(const Kernel2 * k, int x, int y, int z, uint c, WrapMode wm) const
diff --git a/src/nvimage/FloatImage.h b/src/nvimage/FloatImage.h
index 712d1c6..608b440 100644
--- a/src/nvimage/FloatImage.h
+++ b/src/nvimage/FloatImage.h
@@ -81,6 +81,7 @@ namespace nv
         NVIMAGE_API FloatImage * resize(const Filter & filter, uint w, uint h, WrapMode wm, uint alpha) const;
         NVIMAGE_API FloatImage * resize(const Filter & filter, uint w, uint h, uint d, WrapMode wm, uint alpha) const;
 
+        NVIMAGE_API void convolve(const Kernel2 & k, uint c, WrapMode wm);
 
         //NVIMAGE_API FloatImage * downSample(const Kernel1 & filter, WrapMode wm) const;
         //NVIMAGE_API FloatImage * downSample(const Kernel1 & filter, uint w, uint h, WrapMode wm) const;
diff --git a/src/nvtt/TexImage.cpp b/src/nvtt/TexImage.cpp
index b753ef1..884f9c4 100644
--- a/src/nvtt/TexImage.cpp
+++ b/src/nvtt/TexImage.cpp
@@ -1490,6 +1490,16 @@ void TexImage::abs(int channel)
     }
 }
 
+void TexImage::convolve(int channel, int kernelSize, float * kernelData)
+{
+    if (m->image == NULL) return;
+
+    detach();
+
+    Kernel2 k(kernelSize, kernelData);
+    m->image->convolve(k, channel, (FloatImage::WrapMode)m->wrapMode);
+}
+
 /*
 void TexImage::blockLuminanceScale(float scale)
 {
diff --git a/src/nvtt/nvtt.h b/src/nvtt/nvtt.h
index ef13771..709bf48 100644
--- a/src/nvtt/nvtt.h
+++ b/src/nvtt/nvtt.h
@@ -471,6 +471,7 @@ namespace nvtt
         NVTT_API void toLUVW(float range = 1.0f);
         NVTT_API void fromLUVW(float range = 1.0f);
         NVTT_API void abs(int channel);
+        NVTT_API void convolve(int channel, int kernelSize, float * kernelData);
 
         //NVTT_API void blockLuminanceScale(float scale);