Consumer Sony mini-dv camcorders usually deliver severely\nover-sharpened image. I assume this imrpoves the picture when watched\non TV, but looks quite harsh on a good computer screen. Edges develop\nclearly visible halos that can span several pixels. This is due to the\ninternal sharpening algorithm of the camcorder, which normally cannot\nbe skipped.\n\nSmoothing/blurring is the obvious way out, but it took a while for me\nto select the right filter and the right parameters (I work in\nVirtualdub). The camcorder sharpens every field separately - this is\nwhy over-sharpening artifacts are far more pronounced in the vertical\ndirection. So, the first thing to do is to unfold the fields side by\nside (an option of the Deinterlace filter) and apply the filtering to\neach field separately. The corresponding folding filter (another option\nof the same Deinterlace filter) shold follow; the smoothing filter will\nreside between the two.\n\nThe smoother of my choice was General Convolution 3d, which 5x5\nmatrices. This is better than the internal 3x3 convolution, but slower.\nTo define ordirary Gaussian Blurring, one should enter the following\nvalues into the third matrix:\n\n1 4 7 4 1\n4 16 26 16 4\n7 26 41 26 7\n4 16 26 16 4\n1 4 7 4 1\n\nThe central value (41) can be changed to define the "strength" of the\nfilter; the bigger the number, the weaker the smoothing is. I found\nthat the value of 120 works fine.\n\nWhen shooting in the internal 16x9 mode, the horizontal sharpening\nartifacts are even less apparent due to the horizontal squeezing of the\nimage. In this case, the matrix may be modified to only perform 1-d\nvertical blurring:\n\n0 0 7 0 0\n0 0 26 0 0\n0 0 41 0 0\n0 0 26 0 0\n0 0 7 0 0\n\nAfter the processing, the image will appear softer and most of the\nhalos will be gone; however, the worst artifacts cannot be removed\ncompletely.