https://community.intel.com/t5/Intel-Integrated-Performance/Canny-gradient-norm-estimation-L2-or-L1/m-p/826452#M5220 Hello,<BR /><BR />I'm trying to implement Canny edge detector with subpixel precision. AFAIK, there is only usual (non-subpixel) detector in IPP. It accepts x and y gradient components as input and applies hysteresis thresholding based on gradient value.<BR /><BR />The problem is, gradient value is estimated as a sum of absolute values of gradient components: mag = |dx| + |dy|. But all the rest of my algorithm (edge detection is only a small piece) uses more conventional L2 norm to estimate gradient value: mag = sqrt(dx^2 + dy^2). So inconsistencies arise in this case.<BR /><BR />Is there a way to make Canny detector use L2 norm instead of L1, or directly pass gradient value to use?<BR /><BR />Thanks. Fri, 20 Jan 2012 13:35:32 GMT Mikhail_Matrosov 2012-01-20T13:35:32Z https://community.intel.com/t5/Intel-Integrated-Performance/Canny-gradient-norm-estimation-L2-or-L1/m-p/826452#M5220 Hello,<BR /><BR />I'm trying to implement Canny edge detector with subpixel precision. AFAIK, there is only usual (non-subpixel) detector in IPP. It accepts x and y gradient components as input and applies hysteresis thresholding based on gradient value.<BR /><BR />The problem is, gradient value is estimated as a sum of absolute values of gradient components: mag = |dx| + |dy|. But all the rest of my algorithm (edge detection is only a small piece) uses more conventional L2 norm to estimate gradient value: mag = sqrt(dx^2 + dy^2). So inconsistencies arise in this case.<BR /><BR />Is there a way to make Canny detector use L2 norm instead of L1, or directly pass gradient value to use?<BR /><BR />Thanks. Fri, 20 Jan 2012 13:35:32 GMT https://community.intel.com/t5/Intel-Integrated-Performance/Canny-gradient-norm-estimation-L2-or-L1/m-p/826452#M5220 Mikhail_Matrosov 2012-01-20T13:35:32Z