public static void vertical(Kernel1D_F32 kernel, GrayF32 input, GrayF32 output) {
    final float[] dataSrc = input.data;
    final float[] dataDst = output.data;
    final float[] dataKer = kernel.data;

    final int kernelWidth = kernel.getWidth();
    final int offsetL = kernel.getOffset();
    final int offsetR = kernelWidth - offsetL - 1;

    final int imgWidth = output.getWidth();
    final int imgHeight = output.getHeight();

    final int yEnd = imgHeight - offsetR;

    for (int y = 0; y < offsetL; y++) {
      int indexDst = output.startIndex + y * output.stride;
      int i = input.startIndex + y * input.stride;
      final int iEnd = i + imgWidth;

      int kStart = offsetL - y;

      float weight = 0;
      for (int k = kStart; k < kernelWidth; k++) {
        weight += dataKer[k];
      }

      for (; i < iEnd; i++) {
        float total = 0;
        int indexSrc = i - y * input.stride;
        for (int k = kStart; k < kernelWidth; k++, indexSrc += input.stride) {
          total += (dataSrc[indexSrc]) * dataKer[k];
        }
        dataDst[indexDst++] = (total / weight);
      }
    }

    for (int y = yEnd; y < imgHeight; y++) {
      int indexDst = output.startIndex + y * output.stride;
      int i = input.startIndex + y * input.stride;
      final int iEnd = i + imgWidth;

      int kEnd = imgHeight - (y - offsetL);

      float weight = 0;
      for (int k = 0; k < kEnd; k++) {
        weight += dataKer[k];
      }

      for (; i < iEnd; i++) {
        float total = 0;
        int indexSrc = i - offsetL * input.stride;
        for (int k = 0; k < kEnd; k++, indexSrc += input.stride) {
          total += (dataSrc[indexSrc]) * dataKer[k];
        }
        dataDst[indexDst++] = (total / weight);
      }
    }
  }
  public static void horizontal(Kernel1D_F32 kernel, GrayF32 input, GrayF32 output) {
    final float[] dataSrc = input.data;
    final float[] dataDst = output.data;
    final float[] dataKer = kernel.data;

    final int kernelWidth = kernel.getWidth();
    final int offsetL = kernel.getOffset();
    final int offsetR = kernelWidth - offsetL - 1;

    final int width = input.getWidth();
    final int height = input.getHeight();

    for (int i = 0; i < height; i++) {
      int indexDest = output.startIndex + i * output.stride;
      int j = input.startIndex + i * input.stride;
      final int jStart = j;
      int jEnd = j + offsetL;

      for (; j < jEnd; j++) {
        float total = 0;
        float weight = 0;
        int indexSrc = jStart;
        for (int k = kernelWidth - (offsetR + 1 + j - jStart); k < kernelWidth; k++) {
          float w = dataKer[k];
          weight += w;
          total += (dataSrc[indexSrc++]) * w;
        }
        dataDst[indexDest++] = (total / weight);
      }

      j += width - (offsetL + offsetR);
      indexDest += width - (offsetL + offsetR);

      jEnd = jStart + width;
      for (; j < jEnd; j++) {
        float total = 0;
        float weight = 0;
        int indexSrc = j - offsetL;
        final int kEnd = jEnd - indexSrc;

        for (int k = 0; k < kEnd; k++) {
          float w = dataKer[k];
          weight += w;
          total += (dataSrc[indexSrc++]) * w;
        }
        dataDst[indexDest++] = (total / weight);
      }
    }
  }
  public static void convolve(Kernel2D_F32 kernel, GrayF32 input, GrayF32 output) {
    final float[] dataSrc = input.data;
    final float[] dataDst = output.data;
    final float[] dataKer = kernel.data;

    final int kernelWidth = kernel.getWidth();
    final int offsetL = kernel.getOffset();
    final int offsetR = kernelWidth - offsetL - 1;

    final int width = input.getWidth();
    final int height = input.getHeight();

    // convolve across the left and right borders
    for (int y = 0; y < height; y++) {

      int minI = y >= offsetL ? -offsetL : -y;
      int maxI = y < height - offsetR ? offsetR : height - y - 1;

      int indexDst = output.startIndex + y * output.stride;

      for (int x = 0; x < offsetL; x++) {

        float total = 0;
        float weight = 0;

        for (int i = minI; i <= maxI; i++) {
          int indexSrc = input.startIndex + (y + i) * input.stride + x;
          int indexKer = (i + offsetL) * kernelWidth;

          for (int j = -x; j <= offsetR; j++) {
            float w = dataKer[indexKer + j + offsetL];
            weight += w;
            total += (dataSrc[indexSrc + j]) * w;
          }
        }

        dataDst[indexDst++] = (total / weight);
      }

      indexDst = output.startIndex + y * output.stride + width - offsetR;
      for (int x = width - offsetR; x < width; x++) {

        int maxJ = width - x - 1;

        float total = 0;
        float weight = 0;

        for (int i = minI; i <= maxI; i++) {
          int indexSrc = input.startIndex + (y + i) * input.stride + x;
          int indexKer = (i + offsetL) * kernelWidth;

          for (int j = -offsetL; j <= maxJ; j++) {
            float w = dataKer[indexKer + j + offsetL];
            weight += w;
            total += (dataSrc[indexSrc + j]) * w;
          }
        }

        dataDst[indexDst++] = (total / weight);
      }
    }

    // convolve across the top border while avoiding convolving the corners again
    for (int y = 0; y < offsetL; y++) {

      int indexDst = output.startIndex + y * output.stride + offsetL;

      for (int x = offsetL; x < width - offsetR; x++) {

        float total = 0;
        float weight = 0;

        for (int i = -y; i <= offsetR; i++) {
          int indexSrc = input.startIndex + (y + i) * input.stride + x;
          int indexKer = (i + offsetL) * kernelWidth;

          for (int j = -offsetL; j <= offsetR; j++) {
            float w = dataKer[indexKer + j + offsetL];
            weight += w;
            total += (dataSrc[indexSrc + j]) * w;
          }
        }
        dataDst[indexDst++] = (total / weight);
      }
    }

    // convolve across the bottom border
    for (int y = height - offsetR; y < height; y++) {

      int maxI = height - y - 1;
      int indexDst = output.startIndex + y * output.stride + offsetL;

      for (int x = offsetL; x < width - offsetR; x++) {

        float total = 0;
        float weight = 0;

        for (int i = -offsetL; i <= maxI; i++) {
          int indexSrc = input.startIndex + (y + i) * input.stride + x;
          int indexKer = (i + offsetL) * kernelWidth;

          for (int j = -offsetL; j <= offsetR; j++) {
            float w = dataKer[indexKer + j + offsetL];
            weight += w;
            total += (dataSrc[indexSrc + j]) * w;
          }
        }
        dataDst[indexDst++] = (total / weight);
      }
    }
  }