Reputation: 2021
I'm trying to implement a watershed segmentation using Java wrapper for OpenCV.
Here's what I'm doing:
public void watershedSegmentation(Mat image)
{
Mat target = new Mat(image.rows(), image.cols(), CvType.CV_8UC3);
Imgproc.cvtColor(image, target, Imgproc.COLOR_BGRA2RGB);
//Conversion to 8UC1 grayscale image
Mat grayScale = new Mat(image.rows(), image.cols(), CvType.CV_64FC1);
Imgproc.cvtColor(image, grayScale, Imgproc.COLOR_BGR2GRAY);
//Otsu's Threshold, applied to the grayscale image
Imgproc.threshold(grayScale, grayScale, 0, 255, Imgproc.THRESH_OTSU);
//constructing a 3x3 kernel for morphological opening
Mat openingKernel = Mat.ones(3,3, CvType.CV_8U);
Imgproc.morphologyEx(grayScale, grayScale, Imgproc.MORPH_OPEN, openingKernel, new Point(-1,-1), 3);
//dilation operation for extracting the background
Imgproc.dilate(grayScale, grayScale, openingKernel, new Point(-1,-1), 1);
Imgproc.watershed(target, grayScale);
}
Right at the point where I invoke 'watershed', I see an error which reads:
OpenCV Error: Unsupported format or combination of formats (Only 32-bit, 1-channel output images are supported) in cvWatershed, file ..\..\..\..\opencv\modules\imgproc\src\segmentation.cpp, line 151
Exception in thread "main" CvException [org.opencv.core.CvException: cv::Exception: ..\..\..\..\opencv\modules\imgproc\src\segmentation.cpp:151: error: (-210) Only 32-bit, 1-channel output images are supported in function cvWatershed
]
at org.opencv.imgproc.Imgproc.watershed_0(Native Method)
at org.opencv.imgproc.Imgproc.watershed(Imgproc.java:9732)
at segmentation.Segmentation.watershedSegmentation(Segmentation.java:60)
at segmentation.Segmentation.main(Segmentation.java:29)
I get it, OpenCV is hunting for a 32-bit 1-channel file as the output.
I have tried every possible combination:
CvType.CV_32FC1,
CvType.CV_32F,
CvType.CV_32S,
CvType.CV_32SC1,
CvType.CV_8UC1,
CvType.CV_16UC1
... all of them.
The error is resilient. It refuses to go away.
Please help.
Thanks in advance.
Upvotes: 0
Views: 1498
Reputation: 3988
Your grayscale matrix is most probably of type CV_8UC1 even if you created it using CV_64F flag, after going through Imgproc.cvtColor(image, grayScale, Imgproc.COLOR_BGR2GRAY);
watershed
expects the second matrix parameter to be a CV_32F matrix, filled with initial seeds for the regions, see the doc.
[edit] Following the comments, if you just want watershed to run, you can do something like:
Mat seeds = new Mat(image.rows(), image.cols(), CvType.CV_32FC1);
for (int i = 0; i < 10; ++i) {
// initialize 10 random seeds
seeds.at<float>(rand()%image.rows, rand()%image.cols) = i; // translate that into the Java interface
}
Imgproc.watershed(target, seeds);
Upvotes: 3