OpenCv C++ input samples must be floating-point matrix ERROR

While implementing k-nearest neighbour in OpenCV C++, I am using this code:

regionOfInterest = smallt[i](rec);

resize(regionOfInterest, img12, Size(16,16));

img12.convertTo(img12, CV_32FC1);

cout << knearest.find_nearest(img12,5);

I am getting an error

input sample must be a floating point matrix.