a11y: images hide text in ATK tree
Images are provided as children of the pages in the ATK tree, and the actual text is provided by the page itself. So images are always considered "on top of" text, which is annoying for mouse review feature (accessibility functionality) where the user retrieves contextual information from the mouse position: background images shadow the text and text is not recognized at all.
Currently images have no description so they don't actually provide very useful content to ATK -- merely the fact they exist at some location. This might change in the future with tagger PDFs for example (AFAIU), though.
It may be possible to put all images and pages on the ATK MDI layer and force a stacking order, but then background images would be behind the page and in turn, hidden. Also text and images may in reality be stacked in various order.
Setting all text chunks in the ATK tree next to the images and stacking them individually (like actual rendering) would solve this but would have to be done in a way that do not generate thousands of arbitrarily-split text chunks and represent a semantic version of the data, while interleaving images as needed. This is likely to be very complex to do well.
Also it seems that other components are not represented (annotations, media, …) so maybe it could be worth taking this into account on a bigger rewrite operation later when tagged PDFs are there.
For now at our level just disable the image children in the tree as they don't provide real useful information yet and get in the way of mouse review feature.