Colour alone can't separate them
Every edge pixel is a mix of subject and background colour [1]. When both are white — a white mug on a white table, a white shirt on a wall — the mix looks almost the same as either one, and older matting methods fail precisely when foreground and background colours are similar [2]. AI segmentation models such as U²-Net and IS-Net look at shape and context, not just colour, which is why they handle it far better [3][4].
The “colour to alpha” trick for flat art
For logos, scans and clip art on pure white, a simpler method works: treat white as transparent and everything else as opaque, with greys partly transparent. That is the compositing equation run backwards with a known background colour [1]. It's perfect for black line art on paper, wrong for photos, where a white highlight inside the subject would vanish too.
Check the result on a dark background
Leftover white fringes are invisible on white and obvious on black. Preview any cut-out on a dark colour before you publish it — a thin halo usually means the edge alpha wasn't clean.
