0

I am trying to emulate the functionality of alpha compositing in Matlab or to be more specific the compose types CopyOpacity and Over from ImageMagick. First goal is to mask out a region with heavy aliasing edges like the black regions in the images in the set (image set with aliasing edges). This should be done with a compatible gray scale mask were black regions should be eliminated and white should be preserved see gray scale mask. Important is the continuous transition between black and white resulting in a continuous transition between transparent and opaque, see screenshot with transparent regions).

There has been two approaches to compose a foreground FG with and background BF with a gray scale mask CM_mask_blur_alpha but the results were not as expected (see an image set with post-processing with ImageMagick as reference for a pure Matlab procedure).

1) In the first code the gray scale mask is unexpectedly treated as a binary mask resulting in unacceptable aliasing effects at the edges of the formerly black regions (image set for first approach):

FG = uint8(CM_mask_blur_alpha .* FG + (1 - CM_mask_blur_alpha) .* BG);

2) The second approach lead to a visible continuous transition between a foreground and the background but there are aliasing effects from the FG remaining (image set for second approach):

FG = uint8(bsxfun(@times, CM_mask_blur_alpha, FG) + bsxfun(@times, (1 - CM_mask_blur_alpha), BG));

It seems that a one-step approach is not working so I am looking for a two-step approach like in ImageMagick with a masking out of the black regions resulting in an intermediate image with transparency and as second step a compose over of this intermediate image over an background. It is partly a problem setting as in MATLAB: Applying transparent mask over an RGB image and blending with another but there is no gray scale mask and I could not adapt solution parts like the generation of an alpha channel with values derived from the gray mask.

Community
  • 1
  • 1
Günter Bachelier
  • 121
  • 8

1 Answers1

0

It turns out that the problem was not the compositing part but the mask. If the mask is enlarged with a morphological operation (imdilate) with a huge value (se = strel('disk',9)) and then blurred (imgaussfilt) with a relatively large value (like sigma=3) the aliased edges were masked out with a smooth transition. 

whiteImage = 255 * ones(cm_out_h, cm_out_h, 'uint8');
conformal = maketform('custom', 2, 2, [], @conformalInverse_0001, []);
CM_mask = imtransform(whiteImage, conformal, ...,  'FillValues', 0);
% MORPHOLOGY: ENLARGE BLACK WITH imdilate
CM_mask = imcomplement(CM_mask);
se = strel('disk',9);
CM_mask = imdilate(CM_mask,se);
CM_mask = imcomplement(CM_mask);
% BLUR
CM_mask_blur = imgaussfilt(CM_mask, 3);      
% ALPHA
CM_mask_blur_alpha = double(CM_mask_blur)/255;
CM_mask_blur_alpha = im2double(CM_mask_blur_alpha);
Günter Bachelier
  • 121
  • 8