I have read some papers about the topic and searched some MATLAB algorithm. There is one called 'fspecial' in MATLAB. And it could return a Motion filter, when motion is given in number of pixel and angle. I have read 'fspecial' working principles and and it contains many MATLAB built-in function like max, mod, cos, sin, sign, fix, meshgrid, sqrt, abs, find. Is it difficult to implement this MATLAB code in VHDL to obtain a Motion -blur filter.

... This code is portion of 'fspecial' in MATLAB ******
case 'motion' % Motion filter uses bilinear interpolation

 len = max(1,p2);
 half = (len-1)/2;% rotate half length around center
 phi = mod(p3,180)/180*pi;

 cosphi = cos(phi);
 sinphi = sin(phi);
 xsign = sign(cosphi);
 linewdt = 1;

 % define mesh for the half matrix, eps takes care of the right size
 % for 0 & 90 rotation
 sx = fix(half*cosphi + linewdt*xsign - len*eps);
 sy = fix(half*sinphi + linewdt - len*eps);
 [x y] = meshgrid(0:xsign:sx, 0:sy);

 % define shortest distance from a pixel to the rotated line 
 dist2line = (y*cosphi-x*sinphi);% distance perpendicular to the line

 rad = sqrt(x.^2 + y.^2);
 % find points beyond the line's end-point but within the line width
 lastpix = find((rad >= half)&(abs(dist2line)<=linewdt));
 %distance to the line's end-point parallel to the line 
 x2lastpix = half - abs((x(lastpix) + dist2line(lastpix)*sinphi)/cosphi);

 dist2line(lastpix) = sqrt(dist2line(lastpix).^2 + x2lastpix.^2);
 dist2line = linewdt + eps - abs(dist2line);
 dist2line(dist2line<0) = 0;% zero out anything beyond line width

 % unfold half-matrix to the full size
 h = rot90(dist2line,2);
 h(end+(1:end)-1,end+(1:end)-1) = dist2line;
 h = h./(sum(h(:)) + eps*len*len);

 if cosphi>0,
   h = flipud(h);

I wonder, whether I have in correct way to implement a motion filter in VHDL. Maybe there should be an effective way to this task? Or is there already implemented motion -filter in VHDL or Verilog ?

Thank you.


Implementing that function in Verilog or VHDL is far from trivial. Also what I gather this has to be calculated for each pixel. Thus it will need a s[censored]od amount if pipelining and/or parallel processing.

Then you still need a way of getting the pixels into and out of your FPGA. All and all a project which, I won't say can't be done, but I very much suspect is beyond your abilities.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.