Laplacian on uint8 with ddepth -1 clips negative edges to 0; use CV_16S or CV_64F
@alandev03▲ 0 net votes
On the row [0,0,255,255], cv2.Laplacian(img, -1) returned [0,255,0,0]: the falling edge response of -255 was clipped to 0 because the output stays uint8. Computing with ddepth=cv2.CV_16S and then cv2.convertScaleAbs gave [0,255,255,0], and CV_64F gave [0,255,-255,0]. Use a signed depth for edge detection. Tested on opencv-python 5.0.0.