Resizing of large images in 2019a low quality compared to 2014a.

Hi,

I am creating some color mapped plots with imagesc and a custom colormap. The created plot is resized by matlab as it is too big, which is fine, however, the way matlab does the resizing seems to be different between 2014a and 2019a. I have included two images created on the same computer with the exact same data and code: image 1 is created in 2014a, image 2 is created in 2019a. If I zoom into both the 2019a and 2014a images they do look the same, which makes me think that 2019a is corase with its resizing. Things I have attempted are: to make the plots as large as possible to minimize resizing, I tried printing to a vector file, downsampling the datapoints, turning off image smoothing, and I have tried to render with zbuffer, openGL, and painters. I am using a VM and do not have a gpu so I cannot test hardware openGL. My goal is to have the plots look the same between 2014a and 2019a. Or at least remove the diagonal grey lines.

Here is a code snippet from the function that creates the image. It scales down the data to the data limits and then creates the image with imagesc. The data matrix is 65x2880. z is the data matrix. xl, yl, and zl are x axes limits, y axes limits, and data matrix limits respectively.

 %% define the area to plot, axis, scale down the data
    xyLim = [xl(1) xl(2) min(yl) max(yl)];
    ix = find( x >= xyLim(1) & x <= xyLim(2) );
    xa = x(ix);
    iy = find( y >= xyLim(3) & y <= xyLim(4) );
    ya = y(iy);    
    z = z(iy,ix);
    % scale values beyond the limits to the limits
    z(z < zl(1)) = zl(1); 
    z(z > zl(2)) = zl(2);

    % prevent CLIM errors if the z limits are equal or zero
    if all(zl == 0)
        zl = [-0.01 0.01];        
    elseif zl(1) == zl(2)  
        zl(2) = zl(2) + 0.01*abs(zl(2));
    end

    %% make a custom colour bar
    dz = (zl(2) - zl(1)) / (cmapCumSum(cmapIndex+1) - cmapCumSum(cmapIndex) - 2);  % the delta is the diff divided by (range - 1). Excluding the NaN color, so add another -1
    zscal = zl(1)+ dz:dz:zl(2) - dz;
    cbarHandle = subplot('Position', [sa(1)+sa(3)*.95+0.02 ,sa(2)+0.006, sa(3)*0.05, sa(4)]);%vert offset of 0.006 added so colourbar is level with plots
    imagesc([0 1], zl, zscal', [zl(1) zl(2)]);
    colormap(cm);
    title(units, 'Fontsize',fs, 'Fontweight', 'Normal');
    set(cbarHandle, 'xtick',[], 'XTickLabel',[], 'YDir','normal', 'FontSize',fs, 'YaxisLocation','right');

    %% plot the 2D color image/data
    imageAxisH = subplot('Position',[sa(1) sa(2) sa(3)*0.95 sa(4)]);
    imagesc(xa, ya, z, [zl(1) - dz zl(2)]);   % do a trick with the CLim so that NaNs never get confused with real values

Thanks,

K

[Kristen Meyer]

Got though to support and after a bit of time they provided me with a much better workaround than I had done myself. after calling img=imagesc(x,y,z) then you can set the max texture size with img.MaxTextureSize=5000. This greatly improves the software based rendering representation and makes it close to how rendering was done in R2014a. This is not an issue if a gpu exists in the system and hardware rendering is chosen.