Nice! I find font hinting (on both Windows and Ubuntu) so bad that sometimes I struggle to determine whether a space is an inter-character space as opposed to an inter-word space.

The other major rendering issue is the lack of gamma correction. (I imagine using a decoding gamma of 2.5, blending and re-encoding with gamma=1/2.5 should work well, since font rendering is not subject to the rendering intent of the encoding gamma of 1/2.2 for images view in dim surroundings.)

As it is, font stroke weights vary according to their orientation and position. It’s hard to tell what is meant to be bold and what is not. Sometimes in order read a bit of text I have to make a screen snapshot and apply the correction in Photoshop myself (mainly with Adobe’s Acrobat Reader). A black vertical line one pixel wide exactly centred on the boundary of two adjacent pixel columns should have values RGB(193,193,193) in each column.

Obviously this will lighten the overall text somewhat. That’s because the font weight was chosen to look as best as possible on rendering systems with no gamma correction.

It’s kinda odd that most people don’t realise they are working with compressed images. (They are gamma-compressed.) All image arithmetic – scaling, blending, dithering and transparency handling – needs to be done with uncompressed images.