c# - Working with UMat in Emgu wrapper -

i have made application apply hsv filters image , try track specific color. want use umat because need computation of image in gpu , not in cpu instead.

in code below first take image<bgra, byte> , convert umat format hsv color format. however, when split channels 3 channels each channel (ex. the hsvchannels[0] has 3 channels; don't know why happening). can't apply filters because when use inrangeimage method image 1 channel , error form cvinvoke.copy "unknown array format".

the application not best, want first understand how umat image format works , try improve it.

        private umat cvandhsvimage(image<bgra, byte> imgframe, byte lowerhue, byte upperhue, byte lowersat, byte uppersat, byte lowerbright, byte upperbright,         byte erosion = 0, byte dilate = 0, bool hue = false, bool sat = false, bool bright = false, bool invert = false)     {         // first convert input image hsv can change channels         //image<hsv, byte> hsvimage = imgframe.convert<hsv, byte>();         umat hsvimage = new umat();         umat bgrimage = new umat();         cvinvoke.cvtcolor(imgframe, bgrimage, colorconversion.bgra2bgr);         cvinvoke.cvtcolor(bgrimage, hsvimage, colorconversion.bgr2hsv);          // final image returned.         umat resultimage = new umat();         umat resultimageh = new umat();         umat resultimages = new umat();         umat resultimagev = new umat();          umat[] hsvchannels = new umat[3];         hsvchannels = hsvimage.split();          umat img1 = inrangeimage(hsvchannels[0], lowerhue, upperhue);         umat img2 = inrangeimage(hsvchannels[1], lowersat, uppersat);         umat img3 = inrangeimage(hsvchannels[2], lowerbright, upperbright);           #region checkbox color mode         // evaluation of check box each filter. return right image         if (hue)         {             cvinvoke.cvcopy(img1, resultimageh, intptr.zero);         }         if (sat)             //resultimages = img2;             cvinvoke.cvcopy(img2, resultimages, intptr.zero);         if (bright)             //resultimagev = img3;             cvinvoke.cvcopy(img3, resultimages, intptr.zero);          if (hue && !sat && !bright)             resultimage = resultimageh;         if (!hue && sat && !bright)             resultimage = resultimages;         if (!hue && !sat && bright)             resultimage = resultimagev;          if (hue && sat && !bright)         {             cvinvoke.bitwiseand(resultimageh, resultimages, resultimageh);             resultimage = resultimageh;         }          if (hue && !sat && bright)         {             cvinvoke.bitwiseand(resultimageh, resultimagev, resultimageh);             resultimage = resultimageh;         }          if (!hue && sat && bright)         {             cvinvoke.bitwiseand(resultimages, resultimagev, resultimages);             resultimage = resultimages;         }          if (hue && sat && bright)         {             cvinvoke.bitwiseand(resultimageh, resultimages, resultimageh);             cvinvoke.bitwiseand(resultimageh, resultimagev, resultimageh);             resultimage = resultimageh;         }         #endregion          umat grayimage = new umat();          cvinvoke.cvtcolor(resultimage, bgrimage, colorconversion.hsv2bgr);         cvinvoke.cvtcolor(bgrimage, grayimage, colorconversion.bgr2gray);          hsvimage.dispose();         bgrimage.dispose();          return grayimage;      }      /// <summary>     /// method define upper , lower values of channels of hsv image     /// </summary>     /// <param name="hsvimage">image in hsv , byte format</param>     /// <param name="lower">lower value applied</param>     /// <param name="upper">upper value applied</param>     /// <returns>grayscale image applied range</returns>     private umat inrangeimage(umat hsvimage, int lower, int upper)     {         umat resultimage = new umat();         umat lowerborder = new umat(hsvimage.rows, hsvimage.cols, depthtype.cv8u, 3);         umat upperborder = new umat(hsvimage.rows, hsvimage.cols, depthtype.cv8u, 3);          lowerborder.setto(new gray(lower).mcvscalar);         upperborder.setto(new gray(upper).mcvscalar);          cvinvoke.inrange(hsvimage, lowerborder, upperborder, resultimage);          // dispose image due causing memory leaking.         lowerborder.dispose();         upperborder.dispose();          return resultimage;      } 

it looks bug me in emgu.cv, see same behavior.

try using cvinvoke.split() instead so:

private umat cvandhsvimage(image<bgra, byte> imgframe, byte lowerhue, byte upperhue, byte lowersat, byte uppersat, byte lowerbright, byte upperbright,     byte erosion = 0, byte dilate = 0, bool hue = false, bool sat = false, bool bright = false, bool invert = false) {     // <snip>      // final image returned.     umat resultimage = new umat();     umat resultimageh = new umat();     umat resultimages = new umat();     umat resultimagev = new umat();      // << replaced >>     // umat[] hsvchannels = new umat[3];     // hsvchannels = hsvimage.split();     // << >>     vectorofumat hsvchannels = new vectorofumat();     cvinvoke.split(hsvimage, hsvchannels);      // </snip> } 

this method seems give cv_8uc1 umats expected.