Query about 3D Atrous Reconstruction

[simon.guest@…]

Shouldn't the value of x be at least bounded by xdim where the xLim1==xLim2, it is possible to create a value of x that exceeds the x dimension (or is negative). Therefore, oldpos will be incorrect and possibly some random part of the memory and potentially non zero.

Likewise for old row with regard to y bounds.

for(int xoffset=-filterHW; xoffset<=filterHW; xoffset++){
	long x = long(xpos) + spacing*xoffset;

	// Boundary conditions -- assume reflection at boundaries.
	// Use limits as calculated above
		if(xLim1[ypos]!=xLim2[ypos]){
		// if these are equal we will get into an infinite loop
			while((x<xLim1[ypos])||(x>xLim2[ypos])){
				if(x<xLim1[ypos]) x = 2*xLim1[ypos] - x;      
				else if(x>xLim2[ypos]) x = 2*xLim2[ypos] - x;      
			}
		}

	size_t oldpos = oldchan + oldrow + x;

	filterpos++;

	if(isGood[oldpos]) 
		wavelet[pos] -= filter[filterpos]*coeffs[oldpos];
}

[MatthewWhiting]

Address this for upcoming release

[MatthewWhiting]

I have made several changes to the way the wavelet reconstruction handles blank pixels, which has gotten away from the use of parameters like xLim1 etc. The indexing of the bounds is now more robust and memory leaks like those identified in the ticket should now not occur.

Version 1.2 will be released shortly. Closing this ticket.